Hacker News new | past | comments | ask | show | jobs | submit login
Scaleway Stardust Instances (scaleway.com)
124 points by Sami_Lehtinen 61 days ago | hide | past | favorite | 71 comments

Some comparisons of the entry-level offerings at a few providers:

- Stardust, 0.0025€/h: 1 vCPU, 1GB RAM, 10GB disk, unlimited 100Mbps network

- Hetzner, 0.004€/h: 1 vCPU, 2GB RAM, 20GB disk, 20TB 10Gbps network

- Azure, 0.005$/h: 1vCPU, 0.5GB RAM, 4GiB disk, per-byte network

- Amazon Lightsail, ~0.005$/h: 1vCPU, 0.5GB RAM, 20GB disk, 1TB ???bps network

- Amazon Lightsail, ~0.007$/h: 1vCPU, 1GB RAM, 40GB disk, 2TB ???bps network

- DigitalOcean, $0.007/h: 1vCPU, 1GB RAM, 25GB disk, 1TB ???bps network

- OVH, 0.008€/h: 1vCPU, 2GB RAM, 10GB disk, unlimited 100Mbps network

- GCP, 0.009$/h: 0.2vCPU, 0.6GB RAM, no disk, per-byte network

I suppose the big question is how well the CPU performs. Their "price/performance" graph naturally makes them look quite good but it has some issues. The biggest one is that the methodology is entirely missing but there are some smaller clear issues like comparing to Amazon's C6g instance (which costs 13x as much) instead of the price-comparable Lightsail instances.

Lightsail viciously throttles after only a small amount of cpu use. I've not seen anything obvious to compare it to, as it would only be useful for something really occasionally bursty.

> I suppose the big question is how well the CPU performs.

While that's true, Scaleway have an (at best) mixed reputation for IP stability and poor customer support.

eg: https://news.ycombinator.com/item?id=24976899

Agree. I suffered of a IPv6 outage on an existing VPS for various weeks because their network migration broke something and they did not know how to solve it (I actually figured out for them what was the problem after some days, as they looked to have no clue); after ~3 weeks they came out with a "solution": create a new VPS. I had to buy an IPv4 to connect to the VM from outside in the meanwhile, which they never reimbursed me (they offer something, then stepped back). Previous discussion: https://news.ycombinator.com/item?id=22923550

I gave scaleway a try last year during one of their promos, and I recall they immediately double-charged my CC. Not the best first impression, and to me it was the last one as well.

Just to add a counter point of view, I've been a Scaleway customer for about two years now. No problems with my VPS or customer service what so ever.

I have a ticket open for 1 year. They have escalated the issue several times, but no real answer. In the meantime the ticket could be declared invalid because they have changed the implementation but they haven't even done that. Shows that their ticket process is not working. Even worse, the underlying problem is unfixed, they have not updated the kernel for C1 instances for 2.5 years and there is no supported way to run your own one. So they keep charging for an unsupported product (if customers are stupid enough to pay...). They seem to have not enough technical staff and they prioritize new development over support. They are cheap for non-critical use, but I would not trust them for anything business-relevant.

I waited like a year for them to process my GDPR request. Their customer service has been shitty outside of it as well.

First time hearing about Cloud-Mercato. https://vpsranking.com/en/ is something I have seen before and https://github.com/mgutz/vpsbench/wiki/VPS-Hosts allows the results to be replicated.

I haven't used it much lately but LunaNode is a smaller player but with a very solid service offering, with attachable storage &c &c.

Their small m1s instances are $3.5/month or 0.0045$/hr. 1GB ram, 15GB storage, bandwidth is a little harder to understand exactly but "1TB". No affiliation but they've been around & had good low price tiers for a while now, easy to use, worked great, all with APIs for on demand. I think it's all openstack based? Maybe? Love to see.

Lot of great "lowendbox" providers these days. And some not so great ones! Few offer the array of services you expect from a cloud provider (attachable storage, networking, ipv6, &c). Great to see folks who do!

These days I'm mainly on by-thr-month boxes but I appreciate having some short-term load providers I respect & trust.


> Their small m1s instances are $3.5/month or 0.0045$/hr. 1GB ram, 15GB storage, bandwidth is a little harder to understand exactly but "1TB".

Doesn't Hetzner offer a cheaper instance with double the RAM?

Yeah but Hetzner is also 200ms away from the US.

Not quite.

Nuremburg: 85ms from east coast, 156ms from west coast: https://mtr.sh/#ping/8EC5V,MHi1W/nbg.icmp.hetzner.com

Falkenstein: 87ms from east coast, 157ms from west coast: https://mtr.sh/#ping/8EC5V,MHi1W/fsn.icmp.hetzner.com

Helsinki: 107ms from east coast, 163ms from west coast https://mtr.sh/#ping/8EC5V,MHi1W/hel.icmp.hetzner.com

I was rounding a bit. It's 195ms from SF and 209ms from LA here, going over AS6453 Tata (horrible quality network)

Yeah, if they ever managed to expand to a US datacenter they would take a huge bite out of the established hosters there. But, knowing US labor costs, they simply might not be able to do so at their current prices. On the other hand, energy is probably cheaper than in Germany, so there's that...

I didn't mention this, but the provide I mentioned, LunaNode, is only available in (eastern) Canada and France. I don't know a lot about the labor markets of Canada or France, but I have very much wondered whether that plays a part in what offers LunaNode can make. Also not sure what energy costs are in eastern Canada!

On a technical note, seems like ping times from LunaNode to LA are expected to be a bit over 70 ms.

It's OVH.

There used to be a 1€/month VPS at Aruba Cloud (big Italian provider) with 1 vCPU, 1GB RAM, 20GB Disk, 20 TB traffic, WMWare hypervisor... it was just fantastic, but sadly they finally increased it to 2.79€+VAT this summer and I left. Plenty of datacenter locations through Europe though, if Hetzner DE and FI locations are not for you.

I've used Scaleway for several projects across many years, personal experience for those interested:


- often free/unmetered bandwidth - very useful for some projects

- relatively cheap prices

- low downtime in general

- decent support ticket response times


- CPU is generally rather slow, very noticeable compared to mainstream providers like AWS, GC, etc

- little selection of some parameters such as location (only Paris, Amsertdam, Warsaw)

- have had some issues with stopping/starting instances being absurdly slow due to how they move the entire disk around in the past

I think they're a decent choice for side projects that are intended to run at a low cost scale, since the unmetered bandwidth can add up to a lot of savings compared to some competitors (they don't lie about unmetered either - I used to have a $4/month instance that sometimes used >10TB/month, and they never complained), but I likely wouldn't use them for a serious or high-growth startup.

> have had some issues with stopping/starting instances being absurdly slow due to how they move the entire disk around in the past

That is not unusual on other cloud providers too. It is a opimisation they must choose: keep the vdisks close to the CPU(s) that will be running the VM but have the expense of moving it around to put it in the right place on startup, or let the vdisk be hosted anywhere in the network so startup is almost instant but IO performance suffers while running. Most, though presumably not all, choose to optimise for better performace in active use.

Yeah, that seems about right, thank you. But also when I meant 'slow', some times I would wait hours for a disk that was ~200GB, which seemed far beyond what is reasonably expected. It may have been an isolated instance and not the general case there, since my new ones are all much more reasonable now.

One thing to worry about is how long this will actually be available. Scaleway has a shitty history of discontinuing products and pushing the migration work for you.

One example of this is C14. Even though they advertised it as a long-term archiving solution, they gave the product a 5-month discontinuation period after which they just delete your data. If you wanted to start using their new storage solution, you had to do the migration work by hand and then ask them to refund you so that they don't invoice you twice.

Yeah, I decided it was easier to migrate to another company once.

They seem to have the problem that they had a much larger more expensive dedicated box network and need to switch to a modern setup while not only not losing income but gaining for custom cloud design.

Being the customer of a custom cloud has tragic enough costs without paying extra for its build out.

Price/perf of this is hidden by the tiny slice of hardware you get. Equivalent to my Hetzner machine would be 116 Euro/mo. before accounting for the 10x higher peak bandwidth, 2x capacity dedicated NVMe, 4TB of magnetic storage and 100GB of backup space I get for 60 Euro/mo.

I have no idea what "1 vCPU" means in this context but severely doubt it even means "1 hyperthread", unless the physical hardware were something like dual 16 core or single 32 core chips.

Definite pass

> I have no idea what "1 vCPU" means in this context but severely doubt it even means "1 hyperthread" (...)

I don't believe anyone looking into this sort of offering is expecting even or looking for "1 hyperthread". Hell, obviously they aren't looking for performance. I mean, they are marketing this as a node to test scaleway's services and put up small websites. I'm sure that performance-minded customers are planning to run anything performant on a 1vCPU instance with 1GB of RAM.

Scale out, not up! You can run your entire stack on 1gb instances.

Sometimes scaling up is a better choice. Reduced complexity.

I suppose I know where the next wave of vulnerability scans is coming from now.

I have used Scaleway in the past for scanning the internet, and I can say their abuse department is better than most providers. They wanted to know what I was doing and why when someone complained.

The 100Mbps bandwidth cap is not great, but at least it is not 50Mbps like on the Oracle Cloud free instances (which they do not mention anywhere and customer support will play games with you just not to admit it).

Assuming you can sustain this, that's 33TB/month, or 33x more than the [more expensive] AWS lightsail.

It might not be _the best_, but for the price, how is that anything but great?

What's a case where you want to (and more importantly can) burst more than that, but only for a very short time, on such hardware?

> Assuming you can sustain this, that's 33TB/month

That’s missing the point. Most folks don’t fully saturate the pipe ~2.6M seconds a month, but it’s trivially easy to max out a 100Mbps pipe for bursts.

Weird price /performance comparison with hetzner, their cheapest is 2.89€ scaleway is 1.8€, hetzner offers double ram and storage and 1+gbit internet (in my experience usually 2.5gbit)

Is this a tax-inclusive comparison? Hetzner's site shows me prices including VAT while Scaleway shows me prices excluding VAT.

Is it not per of the cpu?

it all seems a bit fishy

there ads for scaleway on the linkedin page on cloud-mercato, "A week ago, Scaleway release the European VM with the lowest price. This week, place to the biggest machine with 448 CPU and 7.6 TB of RAM aka "#baremetall Ultimate Performance"." they only have 1 employee on linkedin, company location paris

the website is owned by a portugese company called balladscale with 2 employees and generates $56,524 in sales (USD).

unless cloud-mercato is just a buggy mess, it feels like its been put up in last minute, there bugs everywhere

after logging in with a computer

they dont even offer a "price score" for hetzner https://imgur.com/a/ymMUTj0

and dont offer a storage performance for scaleway https://imgur.com/a/3qZuxdK

they do offer to a "compute score" comparison with both in it and the hetzner machine is somehow ~3% slower, but that doesnt make up the price difference metrics on scaleway website


Scaleway is owned by Iliad Group. It's a legit subsidiary of a legit multi billion dollar French telecom corporation.


Edit: Ah, the concern is Cloud Mercato, cited in the benchmark graphs.

The fishy parts are about cloud-mercato Not scaleway

I'm using scaleway for more than 3 years. Moved to thm all my websites.

Idont know, the "independent" benchmark site doesn't really work on mobile, n most links go to some kind of broken login Form

But ime scaleway had slower cpu crunching Power and web requests per second than similar hetzner cloud machines, and online. net(parent company) always had routing problems to me which couldn't be resolved in timely manners (1gbit dedicated server only 100mbit arrived at my place)

I had issues so many times with Scaleway where it took insanely long to reboot a server, seemingly getting stuck on some network provisioning, so I ended up ditching them altogether. Might use these boxes for some testing or throwaway stuff though.

Currently happy with DO and Hetzner, and increasingly so with the latter. But Hetzner's lack of firewall is a bit annoying, esp. when using Docker (which interferes with iptables).

> But Hetzner's lack of firewall

the cloud offering doesnt has any which is controllable with the webinterface,but the machines are still firewall protected(mostly for ddos/abuse), but i would just use a firewall vm as entry point and keep the production machines in a private network

u could also use the dedicated servers firewall(the Konsole for dedicated servers has firewall rules in the webinterface, u can attach vswitches to cloud instances)

There is a firewall right? it is just off by default.

right I was thinking of Konsole

It looks interesting. Sounds like they're going to be charging for an entire year of service up front or at least as part of an annual contract, which makes sense. (See: "Stardust will only be the beginning in a year-long partnership.") So this will be 21.99 euros ($26.04 USD) for the year. I wonder if this is an x86 instance or if it's ARM. Pricing would make a lot of sense if it was ARM.

According to the blog post "every service, feature and scenario currently provided via its DEV1-S instances will also be available on Stardust Instances" [0]. The DEV1-S is an AMD Epyc and the technical specs match up as well. So, that seems most likely.

[0] https://blog.scaleway.com/a-star-is-born-as-scaleway-launche...

hmm nope, you are billed per the hour and that's just it ;)

no pre-charging at all

architecture is x86_64

For information, a `cat /proc/cpuinfo` indicates that the instance is running on a AMD EPYC 7281 processor, which has a 2.1GHz base frequency, and a 2.7GHz Turbo freq (https://www.amd.com/en/products/cpu/amd-epyc-7281)

I wonder how this compares to the GCP always free compute instance offering

Seems like a way to upsell their other offerings, one foot inside the door.

i want to know how some providers can advertise "unmetered bandwidth" meaning they don't even log your limits as opposed to "unlimited" which might have an asterisk and many just don't and only provide 1TB or 2TB and the like.

at the scale, does bandwidth cost them or what?

if the costs of such are significant, what would the costs be to someone like youtube?

Took a look at Cloud Mercato reports for compute, and the tl;dr is that Hetzner is so much better value than anything else that it's not even funny. And if you have to use one of the big three, use GCP.

Or use AWS, which is much better in my experience (and not Google).

But yeah, don’t use either. Just go for Hetzner if you know what you are doing.

Nice price for a small vm... If.my math are correct, €1.86 per month (744hours). Not bad...

A month is 365.25/12=30.4375 days on average, disregarding leap seconds.

Some providers use 720 (30 days) hours, some 744 (31days), I have even seen one doing 672 (28 days)...

I really like that even though DigitalOcean charges by the hour, they clearly state a monthly price for the instances and then they cap it at that.

Microsoft uses 365*24/12=730 hours

i haven't able to measure the difference yet due to incomplete bill this month, but removing the IPv4 address can save upto €1

Am I reading this right that there will only be two instances available to begin with?

If so, this is more like a promotional giveaway than a real service. They should have simply said "two lucky customers will be given a tiny VM for almost free".

two instances per account

Except that it has been out of stock for several weeks now, in the Paris region... :(

All good, but they have the most expensive Kubernetes service for small projects!

don't use k8s = issue solved

Your issue, maybe. My issue - not really.

You can run k8s on VMs too. It doesn't require a provider managed service.

True... at the cost of one extra node (at least).

New batch of Stardust instances are now available in Paris and Amsterdam, if you want to give it a try !

But can it run k8s? Haven't been able to get k8s (in any micro flavor) up on any 1 gb VPS yet.

Oracle Cloud had a free tier that's quite good, if you only need something to tinker.

Great, but please note that they probably will raise the prices by 25%-100% in a couple of months. Scaleway's business model is to attract devs and sysadmins by selling cheap, then raise.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact