The challenge for AWS is one lots of incumbents have experienced: they created a market and it's economics
and now they're being attacked by the next generation of market entrants who've structured their businesses to _specifically_ attack those economics.
What's interesting is that challenge can be a really big problem for incumbents, as those economics can form a core (very rigid) part of their operating model; it can make it VERY hard to address without fundamental (read: risky) change to a business. There aren't many examples of incumbent businesses doing it successfully, as it needs a kind of 'self-inflicted disruption' that's very hard to do in large organisations where politics and empire building can make it difficult.
If someone could do Managed NAT Gateway next I'd appreciate it!
they created a market and it's economics and now they're being attacked by the next generation of market entrants who've structured their businesses to _specifically_ attack those economics
Hah, thanks! My comment was fairly blatantly stealing from the book!
It's so interesting from an incumbents internal POV (I saw it a few times during my time at McKinsey) as changing an organisations economics is often the unstoppable force that meets the immovable object of internal politics.
There's a really interesting ongoing example of this in the the UK as 'attacker' banks (e.g. Monzo, Starling) challenge the economics of incumbents. It's not quite the same, as these attackers are removing back-end cost (e.g. branch networks) from an already 'free' product (e.g. retail banking) but it's meant that big banks are looking at their balance sheets and seeing a set of gaping money pits that will require fundamental change in their operating models to be able to get rid of/compete with.
This is a good point but... Innovation without disruption tends to get underlooked, being less dramatic.
Think of the old auto companies over the years. They start off making tractor-like cars. They survive through the cars-as-fashion eras, the internationalisation of manufacturing, etc. If old auto companies emerging from the 80s were new, we'd call it disruptive innovation.
That said, both disruption and innovator's dilemma are real.
The innovator dilemmas also roughly corresponds to stuff early economists wrote about. Peak markets. Markets are great as they grow. When they reach their terminal size (eg most people already own cars), profits go down, stagnation can occur. That stagnation, especially if the market declines in size, leads to crashes and new paradigms eventually emerge. Marxists sometimes take this to a systemic extreme, with "peak capitalism" and derivative concepts. On the conservative side, you'll find these ideas at the heart of austrian business cycle theories and Schumpeter's "creative destruction."
The digital economy is cushioned by tremendous potential for growth, so far. FB, for example, knows that it's not cool anymore. They can just buy whoever is cool.
Slight historical note: most Japanese auto manufacturers started off making motorized bicycles and small utility vehicles, then pivoted up into retail cars.
That reminds me of music industry and the constant buying of smaller labels. Owning distribution is the key and facebook has a massive platform for that.
> The challenge for AWS is one lots of incumbents have experienced: they created a market and it's economics and now they're being attacked by the next generation of market entrants who've structured their businesses to _specifically_ attack those economics.
Absolutely. This exactly what Tesla has been doing with car industry incumbents. For example, the higher specs versions of the Model 3 beat +$100k cars in acceleration, raw power, torque, handling, etc.
Incumbents have been selling performance as a high-ticket price feature for decades. Traditional brands cannot compete on high-performance features against Tesla without cannibalizing their ICE offering.
Too bad they shot themselves in the foot with the cybertruck design. Don't get me wrong I think it's funny/cool that a car with that design is out there, but it just won't be able to eat up the high-end performance truck market even if it has insane torque.
I'm seeing "truck guys" giving a shit about Ford's upcoming all-electric truck in a way they didn't about the cyber truck, except as a curiosity. I think they screwed up the marketing on that in just about every possible way, including the name and the design.
What fraction of Ford's trucks are bought by individual "truck guys" vs. fleet managers? My gut says fleet managers may have more buying power - I can't find any stats online that breaks down F-150 buyers specifically, or trucks in general.
Fleet managers are going to love the idea of paying for electricity instead of fuel and of having extra hauling space. You take your work truck home with you? Great, you're paying to "fuel" it up while it's at home.
I'm not a truck guy but I can imagine most of them are pretty lukewarm about a truck from a company that has never made a truck before, with a design looks more unconventional than all the concept vehicles that never make it to production, that nobody can currently buy.
I expect it will sell like any other Tesla as soon as people get to try it in real life.
Really? Assuming the cyber truck actually ships I think it will be crazy popular. It's a very competitive price for pretty great truck at least on paper. Sure there is a market segment that isn't going to buy anything but an F150 but they probably aren't going to get a electric car anyway. Plus the cybertruck will probably attract as many or more hummer/mall-crawler enthusiasts.
It's an extremely weird looking truck with terrible marketing and a hilarious meme of its window shattering repeatedly. It'll be an incredible uphill battle to sell that thing imo.
Tesla's most recent profits were a staggeringly high $9.22bn.
Ford's most recent profits were a startlingly middle-of-the-road $19.934bn.
The Cybertruck is going to sell to nerds who think they're a handyman, but the eF150 is going to sell like crazy, and Ford has the money to buy up capacity that Tesla can't really match up.
Tesla is selling amazing straight-line performance unlocked by their electric motors, but I wouldn’t rate the Model 3 a better handling car than a 70K Porsche 718.
Much of the mechanicals of handling well still have to be pretty complex even with electric power.
I'm not an expert so I believe you when you say a Cayman/Boxster can handle better. But my understanding is that Tesla's heavy battery pack combined with their dual motor, produces exceptional low center of gravity / traction combination.
I know that most car enthusiasts dismiss Teslas as straight-line acceleration novelty cars, but Tesla is clearly not going after Porsches 718 market. They are going after the German Sedan market where performance has been always their upsell for higher prices (think M-Series or AMG)
> They are going after the German Sedan market where performance has been always their upsell for higher prices (think M-Series or AMG)
The German Sedan market has something Tesla does not nor it will in next 10 years or longer - the build quality. They are just laughingly bad comparing to German trio, in every assembly/build aspect. Once they reach somewhat comparable level of quality (and that's a big if), the trio will have well established EV offering
German luxury sedans are known for comfort and handling, but not build quality.
They break down a lot and are moneypits. Part of the problem is the heavy reliance on plastics that break down with wear, therefore modern German sedans are much less reliable than they used to be.
Another problem is the extreme complexity which also translates to poor reliability.
Another is the high prices of parts. A battery replacement on a BMW costs $300 because the computer system needs to be reprogrammed. A Mercedes fuel pump assembly runs $600 (for a Camry it's $200-300). An Audi headlight assembly is $1100 (for a Camry, it's $250). These are OEM prices.
The high maintenance costs are capitalized as depreciation and are reflected in the resale value.
In my zip code, the private party sale value (accoring to KBB) of a 2012 Honda Accord SE in Good condition with 120K miles is $7K (median). For a 2012 Audi A4 with the same miles and condition, it's $4.8K - basically one of these tricked out new macbook pros with the M1 max chip.
A 2017 Audi A4 with 60K miles sells for 20K - it loses half its value. The 2017 Honda Accord sells for 19K. So it overtakes the A4 in value in year 6.
None of the above is a prediction that an out of warranty Tesla wont also be considered a money pit. Maybe it will -- we don't really have the reliability data yet, and there isn't a robust network of independent repair shops yet, it's all very new. But the German sedans do not constitute a high bar to surpass, the Japanese sedans do.
Build quality does not equate mechanical reliability. I'm talking about the way the cars are built, the chassis, frames, gaps, interior and the rest. Not the engines
Exactly. The Germans are completely willing to over-complicate the shit out of things for 3% more performance out of that system. But when the go to actually build said Rube Goldberg contraption it is dead nuts on and works perfectly so long as you maintain it by the book, and the book might entail replacing an entire cooling system at 100k because why not.
The huge dependence on plastics as a replacement for metal components really does cause a lot of systems to break down.
Yes, maybe they were planned to break down, but you can understand that for the end user, it certainly doesn't appear like a system with high build-quality.
But ignoring that point, yes, the drive-train is excellent. German engines and transmissions are first-rate. However counting on that as giving you an advantage in the world of electric vehicles doesn't make a lot of sense to me.
The low center of mass is definitely an advantage but sportscars are pretty low to the ground anyway. It makes a much bigger difference in SUV size vehicles where a Tesla handles way better than its fossil competitors.
The heavy battery is a disadvantage for handling because heavier things have more inertia. The physics are pretty complicated and I'm not an expert either but if pressed I would point to aerodynamic downforce as completely independent of weight.
We (Cloudflare) have got some things cooking here :)
I'd love to hear more about what problems you're trying to solve/features you'd like to see besides "cheaper" — can you email me at rustam at cloudflare ?
AWS can only have a single NAT gateway per subnet/availability zone(they are usually added in the route table as 0.0.0.0/0). Nat GWs can only scale up so much. If we blow past the limits, then the only option is to use resources from a different subnet. I realize things cannot scale vertically forever, but the fact that one can scale horizontally (by adding more NAT GWs in different subnets) tells me that there could be an architecture that would make this a non-issue to customers.
Also if a NAT Gateway has issues (see the outage on Aug 31st) we, the customers, have to figure out how to route around it.
In Google Cloud you can (easily) add multiple NAT gateways as your requirements grow, while staying in the same subnet. Not sure how far one can go (didn't go past 20 Nat GWs or so). We still have to worry about that (specially since in GCP the number of allowed connections is much smaller), ideally we shouldn't have to worry about this either :)
Azure does not have the same concept because they are bonkers (outgoing traffic goes out of your load balancer (?!))
A single EC2 instance might not cut it. The AWS Managed NAT GW scales up to 45Gbps. They can also support 55k connections to a single destination (multiply that by the number of permutations on your triple - IP addr, destination port, protocol).
If you have single EC2 instance doing the job of a managed NAT, another equivalent EC2 instance is enough to max it out.
You may need a fleet of instances if your requirements are large. Which means that you have a bunch of operational aspects to worry about and the NAT Gateway calculation starts to become more palatable (once you start adding the human cost of maintaining your own, etc).
Pricing is still outrageous though. AWS has economies of scale that we don't.
Yes they did, but they also reportedly have a 30%+ net margin. How is it surprising that other players who are in the position to do so, will attack them on price? While of course offering full API compatibility, which is what challengers have to do.
Do we need board game analogies to explain that some components of AWS are going to get commoditized?
It has been fascinating to watch the price freeze, the collusion, between the three majors in AWS, Azure and Google Cloud. They stopped hatcheting each other on price years ago. The downward price competition used to be very common in the earlier years, they'd frequently undercut one right after another. They like their profitability and oligopoly, so they stopped doing it (among the giant companies only more desperate Oracle continued to aggressively slash at things like egress fees).
Early on, optimizations are everywhere which allow you to pick the low hanging fruit. Ideally, this gets passed onto the consumer.
However, over time, the optimizations become more costly to develop and less of them exist.
Just the other day I got a notification from GCP about new Spot Instances driving prices down by 80% which exceeds their existing preemtible instances.
Similarly with AWS releasing Graviton instances offering better performance and cheaper pricing.
I think egress fees have always been the catch, and I don't think they've seen much price changes over time. So I am excited to see it, but I wonder how much of that is due to the current one directional nature of cloud migration.
Most people are moving to a single cloud. As a result, there probably hasn't been a ton of demand to negotiate the outbound movement. We can debate the merits of the lock in nature, but I don't think that technological improvements really help here. This is just a decision to charge for bandwidth or not.
No. The Netflix mail business was very significantly profitable and the streaming business was losing a ton of money for years. The mail business carried, paid for, the streaming business.
That's because of the the entirely different business model of the disc rental business (first sale doctrine) vs streaming licensing business (you're screwed, the content owners will squeeze you to the wall). The horrible licensing costs of the streaming business is what prompted Netflix to push into production (basically direct those fees equivalent into assets they'd own outright instead of paying all their revenue back out to licensing fees forever).
The horrible streaming licensing cost problem is why Spotify struggles to earn a decent profit despite how much they've grown and having a zillion subscribers. You get no benefit of scale on your margin, because the content owners always squeeze you as you grow.
Spotify is up to $8.6b in revenue and still losing money. Their business has no margin at all, and that's essentially all due to the music licensing costs. That's why they're desperate to push into anything else, other lines of business, where they can not have to pay all their revenue out in licensing fees.
You could also argue that Google tried to reinvent Skype, Slack, Discord, and a million other chat apps, and they cannibalized their own offerings because they were feckless and mercurial.
Yeah, and also cuz they kinda sucked. 1st-gen iMessage, or even old-school Trillian, was loads better than Google's graveyard of shitty chat products.
Google had no overarching chat strategy, just threw gobs of money and different teams at reinventing different spokes of the wheels, never thinking about the cart as a whole.
You needn't use your real name, of course, but for HN to be a community, users need some identity for other users to relate to. Otherwise we may as well have no usernames and no community, and that would be a different kind of forum. https://hn.algolia.com/?sort=byDate&dateRange=all&type=comme...
Also: please don't post unsubstantive and/or flamebait comments to HN. We're trying for a different sort of site here.
Google Talk evolved into Hangouts which then evolved into Chat. It's all one continuous line with a terrible marketing strategy. From what I can tell, Meet seems to be just a confusing way to access Hangouts video chats.
I can still see all of my Talk/Hangouts/Chat history going back years. Removing XMPP sucks, and I was annoyed by that too, but the chat history is still there.
It does seem like CF is coming in and burning down the market instead of capturing part of it. Free is cool for developers but not exactly great for profits.
I can see a long term strategy where the next unicorn starts on CF and eventually pays them money. But it also feels like the big fish will migrate to AWS leaving CF with the cheap clients.
I feel your view of CF is about 4 years old. Combine CF's Cloud strategy with their IT/Security offerings (eg Cloudflare One), they are effectively building a new layer on the internet. Very sticky and hard to replicate unless you cover all bases like Cloudflare. Though, it might usher in a dark age if they are too successful. They could end up owning the internet.
Fair enough. Though I think these cloud products need to be viewed in the context of their other services. The value you from of using these cloud products isn't necessarily their direct feature set. It is that the network activity stays within Cloudflare and when combined with their other products, can't really be done easily with other services.
> So while AWS has 17 ways to run containers and 7 ways to do async message processing, all overlapping and reinforcing and supporting each other, Cloudflare will tend toward introducing singular primitives, stuff them in a box, and try to ship those boxes to as many places as will possibly take them. If they could install Cloudflare on your mobile phone, they would (this gets them dangerously close to being a real life Pied Piper).
I think this statement resonates with me the most - it feels a lot like how I prefer to design systems (ahem, thanks Unix!): simple pieces or types, chained together into systems that are easy to understand, maintain, and scale.
We're still only using Cloudflare's workers and it's integration with caching, but it's getting close to the point where I'd have enough primitives to ship some of the functionality of our system architecture to Cloudflare and gain a net-win for latency and simplicity.
> AWS has 17 ways to run containers and 7 ways to do async message processing, all overlapping and reinforcing and supporting each other, Cloudflare will tend toward introducing singular primitives, stuff them in a box, and try to ship those boxes to as many places as will possibly take them.
Actually AWS also "tend toward introducing singular primitives, stuff them in a box, and try to ship those boxes to as many places as will possibly take them."
It's just that AWS covers such a larger terrotery, that they appear fragmented.
This is why I now almost don't read this type of macro-analysis articles. They themselves lack the overall birds-eye view, because they are usually produced by people with little concrete technical background.
They often is very good at producing analogy, which is very intuitive, but very easily breakdown after moderate amount of details.
At the time we picked CloudFlare, Fly.io wasn't really on ours or anyone else's radar yet. I've been meaning to experiment with Fly.io and Fastly - thanks for the reminder!
Contrary to what the article claims, draws in chess are very common (on the other hand, they're exceedingly rare in Go, and often impossible due to fractional komi).
Sente in Go does correspond to having the initiative, but a move that compels a player into a particular follow-up move should be called a "kikashi" (forcing move).
The exceptions are non-fractional komi, and the exceedingly rare triple ko, which does not technically cause a draw, just an infinite game. Which is generally resolved as a 'draw' by mutual agreement. There are interesting rule variants to exclude the option of infinite games, but they have weird side-effects.
I'd feel confident saying that normal go (19x19 japanese rules with 6.5 komi) does not have draws.
Draws are extremely common in high level play, and statistics don't seem to exist for all levels of play, but I'm willing to guess that it's fairly uncommon across all games of rating 1600 or higher.
According to the Lichess opening explorer[0], across their ~419 million games, only 5.3% ended in a draw. If you change the database from Lichess to Masters however, with a total of 2 million games, about 43% end in a draw.
Anecdotally, I'm rated ~1700 and only 2% of my games were drawn, and most of those were stalemates.
I agree that publicly available large datasets / statistics become fairly rare below the 1600, above that level they are fairly common.
But anecdotally, I once messed around with a bunch of large datasets for the purpose of comparing high-level play to lower ones, and the statistics weren't spectacularly different. Yes, the results are essentially far more random the lower you go (especially below 1800, where play is essentially a lot less accurate), but draws are still fairly common at the 1600 level. If memory serves, top-level games had around two-thirds end in a draw, while at the 1600 level, it was basically down to one third. Not what I would call uncommon, though certainly no longer the dominant result.
> A player whose moves compel the opponent to respond in a local position is said to have sente (先手), meaning they player has the initiative; the opponent is said to have gote (後手). Sente means 'preceding move' (lit: 'before hand'), whereas gote means 'succeeding move' (lit: after hand').
Historically, draw was possible due to both players getting the same amount of points ("jigo"), but when playing under most popular modern rulesets, fractional komi serves as a tiebreaker.
Games can be voided due to a complex ko or superko.
There are modern rulesets with non-fractional komi such as the Ing rules (komi = 8.0) where jigo is possible. But under those rules, in the case of jigo, black wins... making komi effectively the same as 7.5.
For multiple games (e.g.: jubango), a draw can be declared if both players win the same number of games.
This sounds similar to the concept of tempo in chess. A move that comes "with a tempo on a piece" is a move that gains a tempo by attacking that piece.
I'm a huge cloudflare fan. Massive advocate for them but when I do see this talk of them as a new kind of cloud platform I cringe a little. Are we going to under go the same lock-in like experience we've had over the years by using very bespoke closed sourced systems like workers and durable objects. It's one thing to buy into something that does have wide portability like a postgres but much harder to buy into the platforms that aren't open source.
> when I do see this talk of them as a new kind of cloud platform I cringe a little. Are we going to under go the same lock-in like experience we've had over the years
I don’t understand your argument. A relatively small but innovative company is working to provide competition against the big 3 cloud providers … and you cringe?
Even if their service turns out to be more or less a S3 replicate with better pricing (for some applications involving a fixed amount of data that needs to be widely distributed) it’s a win for consumers and innovation
I mean competition overall is a great thing. Personally I wouldn't bemoan disruption of Google et al by Cloudflare.
That said, I remember when I was rooting for Google against Microsoft and Amazom against Walmart. Before my time people rooted for Microsoft against IBM.
Sometimes we want things to become a little more timeless like Linux or HTML where it is democratized and much freer and slower to chamge.
On the one hand, I hear what you're saying. You root for the underdog long enough and they end up becoming the dominant player with the power to match. But this feels like a pretty apples-to-oranges comparison to me.
Cloudflare has to buy, operate, and maintain huge amounts of servers with lots of hard drives, plus all the fiber/copper connecting them across the planet. Linux and HTML are software. They're only "decentralized" in the sense that they don't physically exist anywhere the way that a cloud provider absolutely must.
Cloudflare is still software. We consume these services by writing code after all.
Another example would be postgres. I can rent postgres, including whatever hardware is used to power it, from AWS, GCP or Azure. Or anybody really, like DigitalOcean or Heroku.
My 'postgres' code will run on every vendors service. The same applies to containers.
That is how I understood the comment 'Linux and HTML', something that is standard and universal, that affords portability and let's vendors compete on quality rather than relying on vendor lockin.
Yes, CloudFlare has software, and I think that only further highlights the difference between a complex cloud provider and a piece of software. What good is CloudFlare's software without the vast global network to back it up? Pick a problem, though, and there's probably an open source solution though: CockroachDB for global HA dbs, there's a bunch of containerized drop-in S3 API replacements, etc. But something tying them all together requires a lot of ops work that you don't get through software alone.
Is there something that is fundamental to the cloud that promotes vendor lock-in? I can understand it from operating systems and retailers.
But is there some fundamental obstacle that prevents most cloud services to be delivered by commodity RFC-compliant vendors? Or maybe some glue software layer, that, once you purchase a license, can abstract away the actual provider and make it simply a price decision?
I understand the providers will fight tooth and nail against commoditization, but once the initial wave of innovation and savage competition has passed, do they have a fundamental tool to prevent it?
> But two years from now CloudFlare could be doing the exact same stuff Amazon is doing now, and customers are locked in again, because no source code.
I hear this argument often but it always rings hollow.
A friend had a first gen iPod – when he wanted to switch, he discovered that the music he bought on iTunes couldn't be moved anywhere else because of DRM. That's lock in.
But this morning I was looking at the source code of an app built against the Serverless framework[1] and what I'm seeing is a bog standard WSGI application that uses a library to transform the inbound AWS "proprietary bits" into WSGI[2]. I'm not worried about lock-in there because all API Gateway + Lambda do is "translate an HTTP request into a JSON object and toss it to an app"[3] – what source code am I missing? The underlying Lambda/APIGW code? OK, but do I need it to run it myself? Not really.
Many – most? – AWS products tend towards this analysis. S3 is so locked in that, what, we now have multiple very high quality alternatives that are API compatible?
The real risk of cloud vendor lock in, from where I sit, comes from egregious pricing models that make it cheap to get data in & expensive to push data out. But I'm not sure Cloudflare has the juice to make this play work: egress pricing is essentially free money for AWS, so they've got lots of room to cut costs there – from what I've heard from people who negotiate real bills with AWS, they're very happy to give you discounts there.
This comment doesn't make any sense. I don't see how Cloudflare publishing the source code to their own hosted s3 service would help prevent lockin when an open source alternative to s3 is out there with hdfs. While s3 is a proprietary system, Any programs you write to operate against s3 can also easily be migrated to other object stores (Azure ADLS, Google Object store) with relative ease.
The thing that keeps people locked into s3 are egress/bandwidth cost. Until Cloudflare came along, no hosted object store (Google,Azure, including self hosted HDFS onprem or in the cloud) had economical bandwidth/egress costs.
This is actually one of those instances where I'm not sure how open sourcing a product would make it freer. Don't cloud providers make their dime by what-they-have, i.e. your data, instead of what-they-do (i.e., the source code)? As far as I understand, it's the prices of ingress vs egress that act as the mortar in these particular gardens.
Like if Facebook went full open-source... how does that help, if they retain sole custodianship of my data?
Which means that the REAL question isn't whether they open-source the code (not saying it wouldn't be nice... but it may come with lots of dependencies about their environment that wouldn't be easily replicable elsewhere) but whether their API is open.
And in the case of R2, they mimicked the API for S3. Which is as close to "following a standard" as I think it's possible to get.
Let's be realistic: capitalist organizations should not ever care about source code more than they care about getting money from customers. When you can share code, you do (because "open source" has been a marketing ploy for years now), but when it conflicts with making money, you don't. If they need to lock-in customers to make cash, they will, and if they find themselves a monopoly, they definitely will.
> I don’t understand your argument. A relatively small but innovative company is working to provide competition against the big 3 cloud providers … and you cringe?
Cloudflare is by no means a small hosting provider. By some accounts, cloudflare is world's leading CDN provider by a long margin, far ahead of AWS in this market, and it currently piles up about half a billion dollars in revenue.
Meanwhile, AWS holds 41% of the entire marketspace, with $14.8 billion USD in revenues per quarter. Extrapolating that a bit, $60 billion USD in revenues... $500 million is peanuts compared to this [1].
What Cloudflare is trying to do is remarkable considering what they are up against.
> What Cloudflare is trying to do is remarkable considering what they are up against.
I repeat, Cloudflare is already the world's leading CDN provider, ahead of AWS by a long margin. This is not a David vs Golias story. At most it's a CDN Golias vs a all-in Golias.
It's disingenuous to compare Cloudflare and it's CDN offering to AWS at face value based on gross revenue. AWS offers everything from build pipelines to satellite ground stations, and even provides backup services comprised of a big truck with armed guards.
Cloudflare is impressive and very successful, but it's by no means a small upstart, specially when it serves a market where it eclipse all competitors, including AWS.
In any case, it kind of is a David vs. Goliath. Cloudflare currently employs ~1800 people and has revenues of under a billion dollars. They don't qualify as a large enterprise by anyone's definition. They aren't a 2-man shop but they are very much a David in the broader market. Amazon is an absolute monstrosity in comparison.
I think OP is correct, I'm not sure a judge would say that the "market" here is the entire set of cloud offerings. If the market is CDN, Cloudflare is the current market leader.
I think this is generally how things are seen. For example, in the Apple vs Epic lawsuit, the judge said the market was "mobile gaming", and that in that space Apple was not a monopoly.
Amazon total revenue adds up, but in each of the cloud categories they operate in, are they the leader?
Cloudflare Workers competes with both Lambda and Lambda@Edge. Workers is a general-purpose compute platform that happens to run on the edge; it is not a platform intended to be specific to things that need to run on the edge.
It's probably better said as, "CF Workers competes with Lambda's synchronous use-cases".
Based on what I understand, there are still a few things missing to compete with Lambda's asynchronous use-cases. e.g. Step Functions, 15 min time limits, non-cron events (i.e. events for every CF product), batching events into the same execution, etc. While some of these are technically "not part of Lambda", to compete with Lambda CF needs the ecosystem as well.
Disclosure: 1. I'm an AMZN investor, therefore calling out that the ecosystem is worth keeping in mind. 2. I'm a NET investor, therefore calling out that I'm looking forward to seeing the ecosystem develop :)
I'd say that CloudFront Functions was a closer functional fit (and likely created in response to Cloudflare Workers). Lambda@Edge, despite the name, doesn't actually run at edge locations, but CloudFront Functions does.
Source? Assuming you're talking about 18% of traffic and not percentage of websites, how do you define what counts as traffic in that case? Transfer between AS's? Does internal traffic within AS's count? Does traffic between entities within the same AS count (e.g traffic from one AWS customer to another, or traffic from a Netflix OCA to an ISP?) I'm skeptical of any entities ability to fully measure the throughput of the internet even remotely accurately. The closest estimate you'll likely get is if you're a transit provider able to measure data transfer, and even then you'll be lucky to extrapolate within the correct order of magnitude from that for total global inter-AS traffic.
Why would you think that a company's market cap (not only the relevant portion of the business, but the entire company) is a reasonable marker for how big of a player they are inside of this part of the industry?
Heck, market caps at this point are almost entirely untethered from reality. {cf. Tesla}
Market cap is a reasonable proxy measure for how much money those companies can bring to bear to win the market (especially if losses[1]), should those companies decide that competing is a number one priority. Two examples from Microsoft: XBox (worked) and Windows Phone (failed).
Revenues or profits in the cloud market for each company are mostly a measure of how much they are winning. How much they are spending is a measure of how much they are trying to compete, and the amount they can spend is also dependent on profits in other areas of their respective business.
> Heck, market caps at this point are almost entirely untethered from reality
Most stocks have some basis in reality, and relative value still matters even if you think the whole market is in Lala land. The stocks mentioned are not diamondhand stocks. Variation in valuation is not hitting two orders of magnitude, which is what we have here.
A better measure might be some gross profitability figure for each company that measures how much each company can pump into competing (expenses), but that is hard to calculate, especially for Amazon.
Market cap. has little meaning nowadays especially in tech. It's just a pumped-up number. You could talk about revenue but that's a different discussion.
How does an additional S3 replica with better pricing help the market/innovation except adding one more competitor? And if that's all they end up offering (as per your statement) their cost is to high.
> It's one thing to buy into something that does have wide portability like a postgres but much harder to buy into the platforms that aren't open source.
I tend to feel the same as you - preferring portable solutions that I can host anywhere. However, the reality that we're all building CI/CD pipelines as much as we are actual software nowadays, and moving those from one cloud provider to another is no small feat. Even if you're using some infrastructure-as-code tool to manage all of your resources (e.g. terraform), you can't really `SET TARGET=GCP` and re-run the script (so to speak).
I guess the lesson is: spend as much time picking your infrastructure provider as you do your core technical stack. They're not easy to replace! :-)
Great point about CI/CD pipelines being hard to move between cloud providers. I wish someone will do for CI/CD what Docker/k8s did for cloud deployment and provide a non-proprietary structure that can be easily transferred.
But, depending on your use case, you could also try to describe your build process is some combination of make files and dockerfiles and then just call that from whatever CI you are using.
First time I discovered earthly I found it looked cool, but then I encountered the issue that it needed privileged docker which is not really practical in our setup, as this would require launching one VM per build job (we are using gitlab CI)
Is it still an issue? If yes, any plan to lift this limitation?
I don't see that as an issue right now. They are closed source. But the workers and key/value apis are (so far) either close to native, or very simple in nature. Porting away would be fairly straightforward. It may be a space to watch as more features roll out.
They're smart about this. It's infrastructure lock-in but not at the API/application level, as they are trying to stay as close to "just JavaScript with browser API semantics" as possible. Deno is a project that does this too. If you know service workers and web workers you know Cloudflare Workers. If you know JS OO you know Durable Objects (to a degree).
Think about it, the huge influx of web developers that have been growing up on just using JS. Look at their docs too. It's all very accessible, modern, low friction stuff all while they are selling us their infrastructure. And they communicate in a technical, programmer friendly way as opposed to the business/marketing jargon that we are used to by some of the others.
I think when you say it like this it makes a lot of sense. I'm not a JS dev, that's not my world, but I do understand building primitives for a given audience so if that's their target market makes sense. I just think as they try to battle AWS and explore wider demographics they're going to need to accept some of what that requires. CloudFlare isn't a slick brand like many of the startups around today in JS land. They're playing a different game as a public company so feels like wider adoption is going to require something more.
But saying that, I love when companies push the boundaries and CloudFlare are doing that. Conforming to the norms is just becoming another boring IBM like machine.
What do you mean by "Cloudflare isn't a slick Brand"?
I feel like they're the only cloud company that's been doing any real innovation for the last 5-10 years, and in a very approachable and affordable way.
I am confused. What would you like about CF that needs to be open sourced? Is it the front end? The datacenter operations software? Their algorithms? How would that solve the problem of portability? If there is anything to cringe, it is emotional appeal to OSS without thinking it through. Cloudflare is a massive service provider, not a database engine. OSS has a huge significance in basic building blocks of software - things like openssl lib.
Cloudflare needs to innovate more in order to properly be in a position to do long-term battle with Google and AWS.
Their overhead cost is a concern. As a free service provider to many sites that use them for encryption, they're possibly primarily benefiting (CDN-Wise) from Google's encryption assertions made in Chrome.
A few well-publicized system outages for CloudFlare right now would devastate their entire business model... It's happened.
In order to be independently competitive truly, Cloud Flare would need to probably quickly develop a new mobile phone OS, web browser, and scale their cloud hosting to market prominence very quickly in order to be able to preserve their current market share over the long term, which is a very very steep mountain to climb right now.
It's a very steep mountain to climb, because Google already has the aforementioned things in place, and AWS is firmly embedded with customers that don't want to face huge costs in refactoring apps.
CloudFlare needs to battle Google on many fronts to gain a proper foothold. If I was in leadership, I'd recommend a partnership with a struggling mobile phone company like RIM or Nokia, and possibly with Mozilla on the browser front. Reassuring users about and being committed to upholding personal privacy would be another solid move, and then getting rid of the "utility metered" approach to charging for cloud hosting and introducing simple monthly and annual rates with easier services would likely be ideal moves to ensuring proper growth and market share into the future.
This is the chess game that wins from my perspective... As companies like AWS and Azure develop more and more micro-service and licensing-locked cloud platform apps, it becomes harder and much more costly for those same customers to migrate anywhere else like CloudFlare. This is also why competing with giants is a dangerous game. CloudFlare would need to put a lot on the line to compete.
The smartest hosting customers often stay liquid in terms of which platform they can leverage and migrate to through chess in development, but the process of getting locked into one host platform is now a very real threat. Overall success has always been a chess game to me. Informed and carefully planned strategy, and conservation of resources, always works best.
The SEO impact is negligible at best unless you have it set up to specifically block crawlers (or you just forget about crawlers when configuring rules).
Maybe at some point there were crawlers that assigned spam reputation on a per-IP basis, but so much of the internet these days goes through Cloudflare and other CDNs with shared IP ranges that it would be insane to keep this practice up.
Maybe 2-3 years ago. Pretty sure it was IP based. CF drops you on a shared IP, its hit and miss of you end up on an IP next to a bunch of dodgy sites or not, do a reverse IP lookup to find out what else is running on your IP.
> It would be insane to keep this practice up.
What's the alternative?
Oh yea, did CF ever fix the domain hijacking issue for deleted sites?
Your experience is a bit unusual. We saw measurable improvement from edge caching. Argo routing gave us about 200ms back on TTFB where we thought it was worthwhile. We could of course set up our own edge caching with another provider (we also use Cloudfront a lot), but that doesn’t make Cloudflare bad for providing the same service. Similarly, Cloudflare isn’t bad if they provide a fast DNS alternative to Google’s fast DNS—and the mix of features isn’t identical.
I you don't leverage performance related features of a CDN (mostly cache), it's more a security layer. It won't improve performance until you get your hands dirty or ask a professional to tune it for you (and maybe you did).
A global DNS resolver may decrease performance, for instance it can give poor results on DNS based load balancers.
Interested to know how you assess SEO impact and your findings.
So the interesting thing, back then I think we were willing because of the nascent state of cloud services. We hadn't fully bought into any of this because most were still just buying hardware or renting servers and building their own software. S3 and EC2 were pretty pivotal in the move to this lock-in from a pure infrastructure perspective. Luckily s3 equivalent apis exist on every cloud provider now, its a staple cloud service but I think in 2021 as more things appear, they should be open source first. The open source companies start with that, I think cloud companies should actually open source the tech too.
Honestly you touch one one of the reasons I love Heroku so much. I've never seen a service that manages to do so much of the heavy lifting for me, but at the same time be 0 lock-in. I've helped move 2 apps off Heroku once they hit a point where they needed a bit more operational flexibility and there was zero work to disentangle them from Heroku operationally. Try that with AWS, GCE, or anything else.
> Meanwhile, when people think of "Tier 1" AWS services, its Cloudflare equivalent, Amazon CloudFront, rarely gets any love, and the official AWS Twitter account hasn't tweeted about it in almost a year.
In the last couple years, CloudFront has gone from not really working to actually working very well. Invalidations are now instant, both from the command line and the CLI. You used to be unable to customize response headers, but now you can do that fairly easily.
Maybe they're not publicly talking about it, but they've actually gone and fixed all the major problems.
The article makes no mention of the Cloudflare's enterprise networking tools, and its VPN. Cloudflare is basically in a position to run the internet for most people to buy into it - I have their VPN on my phone and computer, which gets my fast access inside their network. By fronting so many of the world's websites, a lot (maybe a majority?) of my traffic actually flows inside Cloudflare.
Now with Workers, R2, Durable Objects, the server side can move to Cloudflare too. If it makes sense to move servers on the network where the clients are, then this is where they should go.
No, they're both very short term wins for companies, which means they might happen anyway. Cloudflare has demonstrated ethical behaviour so far, but that's not enough to trust a single part with the internet.
Short of the new age web3 stuff, though, not sure what else is a suitable alternative. Competitors to Cloudflare aren't as common because of their gigantic moat — imagine building an org that builds out to hundreds of cities around the world and partners with thousands of network companies.
Definitions of ethical may differ. Shielding far-right sites, cesspits like Kiwi Farms which make it their stated goal to drive people they don't like to suicide, criminals like DDOS vendors, credit card fraud forums, etc. all under the guise of being a "neutral passthrough layer third party" feels incredibly disingenuous to me. The aforementioned people are Cloudflare's customers and Cloudflare hosts their content (yes, sometimes with a short ttl, but the public IP address still terminates at their web servers). They can not be afforded the same leeway that actual internet exchanges are when routing traffic to bad actors.
So no. They may have demonstrated business-friendly behaviour. But ethical? No.
Yes, but my point is that they are by definition not neutral, contrary to what they claim. They get to choose who they do business with and their choices are not ethical.
You should consider them a hosting company, would you consider a hosting company an ethical company if they hosted nazi forums?
With some strawmanning by this definition sleeping is not neutral as I could fight for a cause, or healthcare is not neutral as you might save a nazi's life.
I think the argument for Cloudflare successfully "disrupting" AWS is wrong.
I'd imagine that Cloudflare's master plan isn't to render Big Cloud obsolete (in a Christiansen/MBA case study sense) . They want to become the edge compute equivalent of Big Cloud (AWS/MSFT/GCP).
They've already invested in the physical prescence to do so (similar to the massive fixed cost that it takes to build hyperscale data centers around the world), in a manner that's most valuable to their customers (low latency compute, storage and egress).
This isn't disruption: it's new market creation. It just won't be obvious that's the case for another 10-15 years...
I consider Clouflare to be the least reliable cloud service provider out there. So many CDN and DNS related outages thanks to poor engineering release practices. Considering those are their bread and butter services I wouldn't ever rely on any of their other services.
It is simply incorrect. We have most of our customers on Cloudflare and the larger customers are on enterprise deals. My only criticism to Cloudflare is simply that it is just not as stellar as some of the more expensive alternatives. It is not a high end service but still the right choice for a lot of sites.
When it happens, it breaks a lot of the internet, but "so many" is stretching it - the entire CF network has only gone down a couple of times in the time I've known about them (~6 years).
They extensively use Cloudflare, other than for voice channels which don't use CF's tcp/udp proxy (to minimize ping, since GCP is usually peered better globally).
Cloudflare user for all my services here. I can't remember any downtime ever outside of the couple times where they got massive press over it (because, like, the whole internet broke)
As much as I like to have something else leading this market other than AWS (I hate them for several reasons, but insensitive billing plans, cockpit like interface and lock-in services are the top ones), I'd also hate to see Cloudflare become another AWS.
Are there any tech disruption that will make computing resources affordable for solopreneurs/startups as they once used to be. For the past decade I've seen a very slow gradual decrease in the affordability of cloud computing cost. I trust WASM and WASI will have a huge effect in democratizing the market but I'm not sure yet.
> I've seen a very slow gradual decrease in the affordability of cloud computing cost.
What do you mean by "cloud computing cost"? Digital Ocean will sell you a VPS for $5/month with 1TB bandwidth included. There are tons of hosting providers that offer something similar.
These prices don't seem higher than they were 10 years ago.
Yes but try running a couple of servers with a decent amount of ram say 4GB and we’d notice how the cost goes exponential.
Point being running a couple of servers with a decent ram and a decent amount of storage shouldn’t cost 50$. It should be say, 7$. I know the ask is too much. Just want to see if there’d be any fundamental tech breakthrough to make something like this happen.
A Linode VPS with 4GB RAM costs $20/month, 4 times more then the 1GB instance.
Dedicated VM is $30/month for 4GB, but that's the smallest so can't compare there. But that's only a little bit more than the shared vps, so I'd consider it pretty reasonable.
A few years ago I couldn't find a $5/month option. The cheapest Linode was $10/month.
EDIT checked Wayback Machine...10 years ago, a 512MB RAM Linode cost $19.95/month.
Hetzner cloud has nodes with 4GB RAM for $5.70. For $40 you get a dedicated server 64GB RAM and 2 512GB SSDs and 1 Gbit/s unmetered uplink.
Servers are incredibly affordable. AWS isn't because they don't have to be (giving startups $100k credits and coaching them on how to achieve the strongest lock-in works well for them)
But, while Sandstorm is all about compartmentalizing access to data in a single server, having the document (grain) as its unit, Solid does this with multiple servers (called pods)
Solid is a very different approach from Sandstorm. I wouldn't call it a successor.
Sandstorm gives the user control over both data and compute -- users install apps on their personal server, like installing apps on their phone. Solid focuses on data, specifying standardized storage interfaces and formats, but still expects compute will take place on machines controlled by the developer.
I think Solid's approach is unrealistic. Developers want to choose their storage formats and technologies. Even developers that fully support users controlling their data are not going to want to bind their hands to standardized formats that don't support the unique features that the developer wants to implement, or standardized database interfaces that don't meet the app's specific usage model.
Also, no developer wants to have to access data across the internet from potentially-unreliable servers on the other side of the world.
So I think realistically the code and data have to stay together; the developer has to be able to specify both the code and the data format.
Fair. Sandstorms technical ideas are well represented in Cloudflares product lineup now but not yet it's philosophy. Maybe some of that will change some day. I wasn't aware of Solid, going to check it out!
Yeah, though durable objects are a great idea I do wish they weren't proprietary. I hope they get enough traction to spur the development of a self-hostable FOSS competitor though. (Ideally one that's interoperable with it!)
If it helps, the concept is super-simple and reimplementing such a service won't be hard if anyone tries to make it interoperable with Workers. Miniflare (a dev environment for Workers) implements it in just over 200 loc[0], with the only backend beint Workers KV for data storage (<500 loc if you count that).
> The big 3 clouds are playing Chess, but Cloudflare is playing Go.
I think most lay people don't know the nuances between chess and go and would presume that chess is the more advanced game based on superficial first impressions. Probably not a good metaphor because I don't know the author's opinion on the games and most people will probably see the title and interpret it in opposite ways. Using "3D chess" instead would have been a more clear metaphor.
I think the Go philosophy is probably healthier for an economy overall. I can't say whether that's really what's going on here with Cloudflare specifically, but it's an interesting way of framing the discussion. In particular, the thing that catches my eye is in the "Territory" section of the post, and the idea that in Go it's not "winner take all".
A good Go player won't necessarily beat a less good one by a lot, but will consistently take more territory by the end. Or, as one of my Go strategy books put it: think about a kid cutting a brownie in half to share - they want to give themselves a bit more, but if you're too greedy and try to take a large fraction of it, mom won't let you and you'll end up losing out.
I like the idea that in the economy, good ideas and good companies win more often, in that they get the most marketshare, but not necessarily by a lot.
> think about a kid cutting a brownie in half to share - they want to give themselves a bit more, but if you're too greedy and try to take a large fraction of it, mom won't let you and you'll end up losing out.
We take a slightly different approach in my house. The person that divides the treat, gets last pick.
It's very effective at getting the closest to equal distribution possible.
The only time it falls apart is when I'm not particularly worried, so I haphazardly break the cookie in half and end up with 1/4 for myself.
This is actually a great way to think about it for a number of reasons.
1. Look up what James Hamilton (AWS Distinguished Engineer) has been saying for years about commodity economics disrupting things. It's about the money, stupid.
2. The way AWS has been building out their ecosystem is following a lot of the previous monopolists (Microsoft) playbook. Get other companies to be 'partners' in your ecosystem so they depend on your platform? Check. Training and certification so technologists are tied to your platform? Check, and so on.
3. Amazon and AWS are usually never playing the game people think they are. For example, all the years that people questioned Amazons profits, they were doing their best to hide profits with massive R&D & other investments.
In the case of CloudFlare attacking AWS network/bandwidth pricing, it's worth pointing out that >60% of AWS revenue comes from EC2!!!! S3, and CloudFront is (relatively) small fries.
If that's what they're playing, they're doing a mediocre job of it. They should be forcing Microsoft and Google to rent their cloud services, then using high rents there to force them to sell their own cloud services to Amazon. Not likely to work out for them there.
But, to your point, I'm sure they would if they knew how.
Interesting article, but I have to disagree with the Chess - Go analogy. Pieces in chess do not have a fixed point value. "Knights are worth 3 points" is merely a heuristic that can be moderately useful in an initial assessment of a position...but anyone that plays chess knows that the NETWORK matters. A queen on the wrong side of the board is worth less than a pawn about to promote near the enemy's king; three coordinated pieces are worth more valuable than four isolated pieces.
I'm not actually sure this makes the metaphor less applicable. Network and position matter for both, but the point values in Chess serve to describe the relative value of each piece in addition to that, while the lack of differentiation between pieces in Go means that even more attention must be paid to the network and positions. It is not that network or position don't matter in Chess, but that only network and position matter in Go.
Mmm too much wording for a big-tech overlord free product.
I've used nextdns.io as a "free & limited" and now paying customer.
Get rid of trackers and ads by dns, I get to give them 20usd/year, so I know that their business model should not be to resell my data. There is an affiliate link to give if you are interested.
The missing IAM functionality is also what is preventing myself moving some services to R2. CloudFlare Workers are not 1:1 with AWS Lambda either, yet they have seen significant improvements, which likely continue to accommodate for more use cases. I suspect R2 will see similar improvements.
AWS having high egress fees is the moat around their business. If AWS respond by lowering egress costs then they are opening the fort.
The differences between web2 disruption and web3 disruption strategy games is like the difference between robin hood and bladerunner.
Google went public @ $20bn and the papers were full of stories about Googlers getting filthy rich. Now bloggers casually comment on scrappy $10bn incumbents and the possibility of integrating a literal currency mint. web 4 is gonna be a bastard.
I wish douglas adams was still around to explain this all to us.
I'm thinking there's an interesting parallel between our browser-based p2p project [1] and cloudflare workers / DurableObjects. Instead of DurableObjects, we got HashedObjects [2], and instead of workers running on an edge network somewhere, we got in-browser p2p nodes running a browser-to-browser mesh network.
In general, what they do with infra, we do with cryptography & datatypes.
Can someone share a link that describes Clay Christensen's thought or analysis on his management style? Watching Prince explain the Innovator's Dilemma piqued my interest
I believe some reviewers of his book say that the book is his HBR writings organized into a book. In case you're not aware there is the actual book Clay wrote as well - https://en.wikipedia.org/wiki/The_Innovator%27s_Dilemma
Odd that this article suggests that Intel ignored a new technology until it was too late in the Apple case, when the article that they link to back that claim argues for the other style of disruption---low-end product eventually claws up market share and performance to compete with the high-end.
The rest of the article seems reasonable, but IMHO and many other's opinions is that the Intel/Apple/ARM thing is classic disruption from the low end.
The book “On China” by Henry Kissinger makes almost that exact argument. Whatever your opinion on Kissinger, he opened relations with China and definitely has an interest viewpoint.
It’s a bit of both. Lots of countries have low wages and lax safety protocols, but they haven’t been able to hit double digit gdp growth for decades because of that alone.
AWS doesn't need to copy Cloudflare. It already has literally everything Cloudflare does in their catalog already. In spite of this Cloudflare is still attracting customers at premium prices.
Cloudflare is still in growth mode: They're losing money hand-over-fist. AWS, on the other hand, is a money-printing machine.
Personally, I don't trust Cloudflare until they achieve profitability. They're going to have to raise their rates one day, and alienate the majority of their customers.
Probably not really losing money, depends on the cost of bandwidth. Since CF has purchased tons of pipes, it doesn't cost them that much to feed slightly more traffic into it.
I'm not really worried about the cost of traffic. According to their latest 10-Q [0] they earned 152 million and spent 187 million for a net loss of 35 million. About 76 million of that (40.6%) went to sales, so they're certainly not spending it on the pipes. But despite a large increase in sales spending they're less profitable than they were during the same period a year earlier. They're losing $1.23 for every dollar they earn.
This is often a business decision. Cloudflare's bandwidth is free, and with smart tiered caching my operation serves 6TB a month while only paying out 125gb of AWS egress (with extremely hot files).
Wait can you please elaborate more? I have a site that gets a ton of traffic and my biggest bill is AWS cloudfront bandwidth... how can i reduce this using cloudflare?
Well, the currently most used paradigm for building web is that you see the edge servers as your classic web servers and then see the cloud as a service layer. Good for security and scaling. Maybe you can achieve the same thing within AWS.
Well, they are competing for same market, so whatever game it is, it's the same game. Perhaps it's Fluxx [1], a game where you can change the rules. Perhaps, from the authors perspective it is fight to the death, and AWS is infantry lines against Cloudflare guerrilla warfare.
But isn't it simply that Cloudflare is following the disruptors handbook? And therefore isn't AWS most likely fully aware of what Cloudflare is up to and what the avenues (revenues) for attack are, rather than bumbling around playing the wrong game?
A different problem is that, at least with federal agencies, Cloudflare has a BAD name. Like unbelievably bad. They do have a FedRAMP offering as of this year..
But I've been on calls with agencies. Dept heads, executive yuck-de-yucks. And we've gotten, "Are you using Cloudflare?" We don't, and say so. Resoundingly, we get "GOOD"
We have no clue what the story and history is there. It's bad for sure. And nobody will answer why.
On the commercial end, this makes sense. But damn, egress from the majors suck. But that's roach motel computing...
I'd wager that dept heads and executive yuck-de-yucks by and large only know what they heard from other dept heads and executive yuck-de-yucks, which is that Cloudflare didn't buy into the censorship-by-boardroom-committee plans of the two American political parties over the past few years.
Cloudflare’s priority is growth. They intentionally take on customer risk and technical risk to try to maximize growth.
As a result they incline toward hosting whoever wants to use them, and moving fast and breaking things. Neither of these align with typical federal govt approach to IT infrastructure, which emphasizes reliability and avoiding known risk.
It’s just a big personality mismatch, and there’s no reason for either to resolve it. Cloudflare doesn’t need the feds, and the feds don’t need Cloudflare, at least not commercially.
probably the fact Cloudflare doesn't just shut down sites at the governments request. I'd also imagine a good chunk of them have been wined and dined by Oracle, Google, Microsoft, and AWS lobbyists to think that Cloudflare is bad. I doubt non-technical federal agency heads are double checking what they are being told
Cloudflare is a CDN. Nobody is going to use them to store their data even if they are cheaper. If customers cared about price they are already using B2 and B2 is still cheaper than R2.
Cloudflare is not eating anyone. They are just trying to expand their TAM. Cloudflare has always been very good at engineering marketing, and R2 is another masterclass but it will never eat S3
> Cloudflare is a CDN. Everyone would readily consider using them to store their data since they're cheaper. Customers that care about price may have cheaper options, but Cloudflare has excellent engineering marketing.
> Cloudflare will be eating everyone. They are trying to expand their TAM, and R2 is a masterclass.
Figured I'd throw another overconfident unsubstantiated claim into the mix. I was even able to use the same exact points to argue the opposite position.
If in 2-3-5 yrs it's proven to be both durable & highly available (I'm looking at your maintenance windows, B2) I don't see a lot of reasons not to move away from S3 as this should be plug & play at this point.
There will be edge cases for highly regulated businesses of course, but for an average startup why not?
Akamai netstorage was/is expensive, requires a contract and interacting with inept, overpaid and rather useless sales people and sales engineering that insist on coming to your office to yap about their awesomeness a-la IBM, and is a part of CDN which is also expensive and also requires a contract with more sales people and sales engineers that insist on coming to your offices to yap about their awesomeness.
I cannot wait until someone finally puts Akamai out of its misery -- they stopped being an innovative company in 2000s.
Pretty nice for Cloudflare that their closest competitor makes very little effort to gain customers, and actively avoids the hobbyist set which is key for mindshare.
Not the first feature to fall in that category. I do not think it's that big for the industry overall. Cloudflare is better at PR / more visible than Akamai.
It might be interesting for markets where Akamai is not really competing (low budget?). S3 compatible API also is a plus.
I think much of the same could have been said for Heroku and its ecosystem. They tried a few critical services and plugins for everything else. It works great for some things, but not the enterprisey ones that are actually the profit cows.
To win this game, surrounding territory is not enough. You have to go for the king.
hey author here! thanks for posting this, i guess my original title wasn't HNbait enough huh :)
happy to take any questions, and yes acknowledged that I dont follow pro chess at all, keeping it in there as a reverse shibboleth and a reminder that i'm just a rando guy on the internet who can be wrong
You brought many fascinating ideas to the table with this article. As someone who's seeing this for the first time, and adjusting to the paradigm shift you laid out, I'm curious to know what was the context which led you to write this article. What sort of ideas, resources and events helped you connect the dots and express CloudFlare's plan in this way?
thanks! I worked at Netlify and AWS before my current job at Temporal, so:
- I've spent quite some time thinking about how "new clouds" compete with Amazon
- I've seen Netlify argue (with mixed results) that its users should not put Cloudflare in front of Netlify
- I've had casual chats with Rita and James (mentioned in the article) that got me really thinking about what their strategy is. I've had "eating the cloud from outside in" since the start of the year - the Go analogy only came that weekend when I finally sat down to write this thing and R2 was just freshly out.
- I've listened closely to all of Ben Thompson's stuff
You should look up a game called Hive. I like to think of it as a "modern" chess. Games typically last 10-30 minutes and has just as much complexity and strategy (in my opinion) as chess.
Then everyone who had to run in AWS now has the option to think about running outside. Which establishes a new market sector and puts enormous (?) downward pressure on the price of some internal services.
Cloudflare is not worried about this, they want that, because it would open market access to a lot of juicy potential clients, who are already cloud ready but AWS locked in.
Plus they have this shot, they try to make this count, to get traction. If AWS moves now it'll be attributed to them. At that point they win by default. (At least that's the theory :))
Every data processing workload is basically S3 based. Hadoop in the cloud is nothing more than X on HDFS on S3.
But egress fees apply for everything, not just S3.
Currently cross-cloud or multi-cloud orchestration and/or scheduling makes no sense, because egress fees just make it uneconomical (in most cases). The lower the fee gets the better the numbers will look like for mixing and marching services from providers.
B2 pricing is 0.5 cents/GB/mo, R2 is 1.5 cents/GB/mo.
B2 egress is 1 cents/GB/mo with 1GB/day free, R2 is free.
If your cloud storage is for backups, B2 is likely to be less expensive because backups are rarely downloaded and their 1GB/day of free egress is enough to do backup maintenance to optimize storage.
Cloudflare's CDN can proxy a B2 bucket to get free egress and maybe faster downloads (haven't needed it myself):
Do remember that Cloudlfare's CDN is not meant to serve non-webpage content. They outline it in their ToS; section 2.8 here: https://www.cloudflare.com/terms/
It is unlikely that this same restriction would apply to R2.
I just checked, and Scaleway's object store pricing is 0.01 euros/GB/mo (0.012 cents/GB/mo) vs Backblaze at 0.005 cents/GB/mo, so Scaleway is 2.4x more expensive at today's exchange rate: 0.012/0.005 = 2.40
They do have C14 Glacier-like object storage for less, but it also has a unique workflow where data is uploaded to a temporary area, archived 7 days later, has a unique archive id that has to be kept and then used to restore archives back into the temporary area.
I tried to signup for Scaleway's free 75GB/mo account just now, to run some performance tests with HashBackup (I'm the author), but they won't allow creating a bucket until credit card details are entered.
By contrast, Backblaze lets you create a new account w/o credit card info, within the free 10GB limit. You only need a credit card to go past 10GB. They have a big button to click to run free proof-of-concept tests that exceed 10GB.
The problem with B2 is the API request costs can easily bring it over 1.5 cents per GB per month. If R2 can keep to free egress and free (or at least the cheapest) API requests, it will blow all other competitors out of the water. The only provider who provides free S3 compatible with free egress and free API calls is Linode at 2 cents per GB per month. The downside with Linode is your S3 is limited to one region. For now though they are an amazing choice because I can have cheap S3 with unlimited egress in the same region as my managed k8s, also with unlimited egress. The main thing that stresses me out with Linode is having to manage my own SQL database...
This reads as though the entire chess vs go conceit was meant only to bait eyes into making it all the way down to the last paragraph, which jumps the shark by dignifying Web3 nonsense.
Amazon prides itself on the "race to zero cost" as a way to beat competitors. AWS will release a service with feature-parity at the same price and customers will default to that. so cloudflare is learning to play checkers poorly.
Counterpoint: Cloudflare is a poorly run company that is well known for paying cheap. They don't have the clout to be a successful slavedriver like Amazon, so unless they shape up, they're not getting on the level of Microsoft/Amazon/Google.
The challenge for AWS is one lots of incumbents have experienced: they created a market and it's economics and now they're being attacked by the next generation of market entrants who've structured their businesses to _specifically_ attack those economics.
What's interesting is that challenge can be a really big problem for incumbents, as those economics can form a core (very rigid) part of their operating model; it can make it VERY hard to address without fundamental (read: risky) change to a business. There aren't many examples of incumbent businesses doing it successfully, as it needs a kind of 'self-inflicted disruption' that's very hard to do in large organisations where politics and empire building can make it difficult.
If someone could do Managed NAT Gateway next I'd appreciate it!