Scaleway (subsidiary of Online.net) offers even better value/performance deal than DO/Amazon/Linode. Their offering starts at €2.99/month for 2GB RAM, 50GB SSD, 200mbit unmetered cloud server [0]
The 200Mb/s unmetered pricing is really incredible.
I have a server moving 20+TiB up and 20+TiB down per month for 2.99€.
If I'm reading this right with EC2 pricing I'd be paying 1,750€/month just for the bandwith!
This comment actually got me to check them out, so right now I'm deploying 2 bare metal servers and benchmarking them against my DO boxen with interesting results. The SW boxen are outperforming the DO ones at times over 8x in disk IO, and connection to remote servers using vpsbench
Well, you're obviously costing them money. (I will say that I personally think it's a bit abusive of you - I mean 20 GB would be one thing, but "20+ TiB"... I mean it's like going to an all you can eat buffet and eating 70 whole lobsters. 7 is one thing - but 70?) At least you're giving them free press here. :)
On that note - how are latency and CPU access? Can you run a fast, performant site on it? (If you have actual numbers it would be great, such as ms to serve first page, for example, on an actual install.) Maybe some people here can help subsidize your excesses by hosting their tiny web apps on it... ;) (And knowing that if their site ever does blow up - suddenly spike in traffic - they'll be covered without huge bills.)
AWS and Google Cloud want people to think that, so that they can keep selling people McDonalds and charging Michelin prices for it.
Market-rate quality bandwidth is closer to $0.005-$0.015/GB. That's with peering to eyeball ISPs like Comcast and everything. Don't believe me, go to Voxility and crunch the numbers yourself.
Their marketing guys say about it what they're going to say. But under the hood, you tell me if you're getting something different:
Oh BTW, DigitalOcean's site just resolves to a Cloudflare controlled IP. If they can't successfully run their own servers on their own network, I won't either. You can have a different opinion here, but that's mine. I like to see my ISPs eating their own dog food.
That said, they are a business, and it's their job to make their bandwidth policies clear and enforce them. Either they're ok with a few users using tons of bandwidth or they can't be bothered to put in place a clear policy. Either way, it's not the user's problem.
How much bandwidth do you provide? What is the transfer limit?
Your server comes with between 200Mbit/s and 800Mbit/s
of internet bandwidth depending of your server. Checkout
the pricing page to find the exact value. The internal
bandwidth is only limited by the speed of your network card.
There is no transfer limit, transfer is unlimited.
That's the bandwidth policy. I've never had them crack down on me using the bandwidth I'm paying for, never heard anyone else say they've had a problem either
> I like to see my ISPs eating their own dog food.
I have to disagree in part here - No matter what is going on with the ISPs networking etc I want to be able to see status updates. Hosting off-network is great for this.
In DigitalOcean's case they can of course use multiple of their own regions I suppose, but if Cloudflare does the job the box is ticked for me
I would hope it's not abusive of me, I think I'm using around 136Mb/s of the promised 200. It's a Tor relay so it happily takes all the bandwidth it can get.
And I sure am happy to give them free press, it is pretty incredible :)
The CPU on the other hand isn't the beefiest (there are comparative benchmarks floating around), the advantage is probably going to be the predictable latency since it (almost) all runs on bare metal.
I don't understand your numbers. 1 month in seconds is 2,629,743 seconds [1] and at 136 MB/s that is 357 TiB [2] - almost 10 times what you quoted. Are you really averaging 136 Mb/s day-in, day-out?
Also, I'd like to understand this:
>the advantage is probably going to be the predictable latency since it (almost) all runs on bare metal.
Surely you get a VM don't you? How can it run on bare metal while giving you full access to your image? What did you mean by this bare metal remark . . . Thanks!
>I don't understand your numbers. 1 month in seconds is 2,629,743 seconds [1] and at 136 MB/s that is 357 TiB [2] - almost 10 times what you quoted
Almost exactly 8 times, as a matter of fact :)
I use Mb/s for megabits/s and not megabytes, so that comes out to a bit more than 40TiB, 20 up and 20 down.
(The 136Mb/s I estimated was a big too high, actually. Here is my bandwidth chart if you're interested: https://i.imgur.com/l1rCeHC.png)
>Surely you get a VM don't you? How can it run on bare metal while giving you full access to your image? What did you mean by this bare metal remark . . . Thanks!
I believe they have their own custom hardware that they manage directly, they spin one up and you run bare metal on one of these with some local SSDs attached.
With the exception of, I think, the VPS which is KVM.
Thanks. Right, I just let Google do the conversion (per my footnotes) and didn't think about the bit/byte thing - ooops.
Do you run a web server on yours? Can you check locally what your actual latency is for curl-ing (or wgeting) a page? I mean the actual timestamp difference from your local machine, such as your laptop or desktop. I don't intend to have a lot of traffic but I really, really hate latency. thanks for all this info btw.
Do they require you to send in your documents (drivers license, utility bill, etc)? OVH required I send mine in to pay my next invoice after a few months of paying on time with no abuse complaints (nor doing anything that would generate such). I dumped them in the end :/
(OTOH, I've used one of these type of providers listed in the post for a few years without them demanding my papers so maybe it's just an OVH thing?)
I ended up on Scaleway after OVH led me on with this nonsense as well. For me however, it was when I first signed up. After getting the documents in order (which actually wasn't easy, as it was for my startup and we didn't have much in the way of paperwork), they just had me sitting there waiting and waiting for the server to actually be provisioned. It was very painful. I cancelled - at least the refund process went smoothly.
Ended up on Scaleway, no paperwork, no nonsense, no problems. Love it. I also have a $5 Vultr server and am very happy with it, though it's just not as good value for the price, for my needs.
OVH didn't require hardly anything from me, but they did have a (temporarily?) broken provisioning system where my server that I paid for vanished about 8 hours later, and my billing information was also erased - but my Paypal bill shows I paid.
OVH's customer service and support was beyond useless. I just gifted them the €14 and told them to fuck off since it was clear I would spend hours just for the hope of getting a refund.
Other people on the internet have told their horror stories about OVH, but I chose to ignore those. I guess sometimes the stories are true.
Tried that, but it was going to take 30 days and I was going to have to prove I had attempted to get the issue resolved with OVH first. Of course I considered doing that until after the first two rounds of emails with OVH support where it seemed they were either answering another user or had a random script program sending responses.
I was annoyed enough that even trying to explain the support communication failures would take more energy than €14 was worth to me.
Interesting, I have been running an OVH vps for over a year without that. Until I lost my 2 factor auth backup keys and needed them to disable it. Which seems reasonable.
I've had a quick look at your site because I'm currently using DO and Vultr with a PeerVPN[0] setup to connect the individual hosts - which brings me to my question:
Is it just a static website or do these instances communicate with each other?
If you do, I'd like to ask if you use a VPN to connect them (and which one)?
I'm relatively happy with PeerVPN (with one or two rather minor annoyances that most likely can be solved if I actually spent time on them) but am always curious about alternatives.
The site is mostly static, expect the server-side Python code to handle the submission of blog comments. I use no VPN but have automated rsync runs, over SSH, to sync data files periodically. A description of my setup is at: http://blog.zorinaq.com/release-of-hablog-and-new-design/
> I like Vultr because they accept Bitcoin payments.
This. I've been a loyal Linode user for years but recently gone with Vultr for side projects because they accept Bitcoin and had a $5 plan long before.
As far as the CPU goes, try re-running sysbench with --num-threads=2. Since you get two cores, if your app can take advantage of that, the speed is about the same as the others.
I'm sure that's great for Europeans, but having US datacenters and USD pricing is a major feature of DO/Linode. They don't seem to be competitors so much as serving different regions. Latency to Europe is quite significant.
I had one planned maintenance disruption using their API for a few minutes as well as about 30minutes disruption yesterday solely with their API. I told them about the latter via Twitter. I am not sure if they found this error themselves or if my Twitter post helped them. The status/incident-ticket page was updated after my Twitter post. I don't know how much of these 30 minutes was fixing the issue or just discovering my Tweet. Once they contacted me on Twitter it was fixed in < 5 minutes.
As jsnathan mentioned below, Scaleway's offerings are all multi-core. Sysbench defaults to using only one thread - you need to tell it to use more. (In your case, use --num-threads 4)
That's great and very similar to OVH prices. I host my main site and documentation section and needed something basic without bandwidth restrictions. I never exceed more than 50gb pm anyway but it's a nice to have.
Ovh fits the bill but next time I'm looking, scaleway is definitely getting my business next.
That's really not that much traffic, so I don't think any of them will not be able to serve up so few hits. One of our servers is running on a $40 Linode and served up 186,698 unique visitors, 6,837,460 requests, and 552G of data last week alone and we had no issues with performance (in terms of bandwidth, CPU, memory usage, or disk usage). That Linode is a tad bit beefier than the $5 one, but I doubt you'll run into any issues with 50k visitors on a Ghost blog (which is, from what I saw last year while playing with it, fairly performant and optimized).
If you are going the VPS route, just make sure the provider allows for resizing your instance in case you do end up needing a larger instance.
KVM for the VCS, otherwise bare metal. See their faq.
It gets weird if you would like to pass additional kernel parameters during boot. My hope right now is that this might be possible via their API but not their website.
You can select a precompiled kernel from which you want to boot your system. If you need other kernel features, you tell them so they add these.
Machines boot via ipxi and fetch the kernel. Once kernel is loaded the system is handed off to your rootfs.
You can install Windows on a bare metal machine, if you install a proper hypervisor. I tried it using KVM with Proxmox.
I did not try nested KVM.
I have read something/somewhere about Ubuntu MAAS, but I am not sure how this fits in.
[0] https://www.scaleway.com/pricing/