Scaleway – NVMe SSD Servers and Hot Snapshot | Hacker News Reader

Scaleway – NVMe SSD Servers and Hot Snapshot

blog.online.net

130 pointsolragon60 comments

...providing that they can actually and predictably supply the servers.

My experience with Scaleway was essentially this: they advertise that you can have virtual servers with up to 10 150GB SSD volumes, so I signed up and created a virtual server with two 150GB volumes. After several attempts over three days, the server wouldn't start at all; it never actually started.

Scaleway's support told me that "it's because there are no available nodes matching your configuration." and "If our stock is low then there is more chance that the only free nodes are "default" ones, with 200GB available only". The solution, according to their support, would be to "keep trying to start the server until it works".

So, they'll sell you a vServer but can't guarantee when you'll be able to spin up the machine.

For me, it was a terrible experience, and Scaleway's "customer support" forums will give you a good grasp of what you're in for if you buy their service.

I also found that in addition to that even when the server was available I had a bunch of problems even getting it to start. Their customised images just seemed to not work that day.

Once I finally got the image started I then attempted a basic update && dist-upgrade and find myself looking at a bunch of errors, they had reconfigured the entirity of their image to point at their own caches, but thoses caches weren't there.

I finally contacted support and their answer was that this was my fault for choosing an old Ubunutu instance (I hadn't it was new ish). I went to the forums and all I found was people with major stability issues, problems with availability and broken images. They are cheap but awful both in basic delivery and in customer support. It is not worth wasting your time with. It cost me 5 cent just to mess around and not get anywhere and that was mostly a ridiculous bill after they wasted my time so incredibly and I never got a working box.

This is not a company worth dealing with.

hello, my personal website is running on a scaleway C1 (dirt cheap baremetal arm) since 3 oct 2015. There were some network issues the first months. No problem since. My main administrative task is: apt-get update/upgrade every week and backup of my site every couple of month.

  > apt-get update/upgrade every week

Have a look at: https://wiki.debian.org/UnattendedUpgrades

Looks like that's a physical, not a virtual.

I use the same. I don't have any uptime requirements because it's only for freetime and experimental use. But I don't think there has been any outage (knocking on wood)

If you use their Ubuntu image it should have unattended upgrades enabled already. Still, you need to watch it. Ubuntu is not very good at restarting daemons after a library has been upgraded. dhclient is regularly affected. And libc is blocked from automatic updates IIRC. The kernel anyway, because this is a diskless setup.

I like the product (baremetal ARM). But I fear it's dead. They used to upgrade the kernel frequently, but now it appears that no one is working on it anymore. Of course don't try to run anything that requires performance or availability on a 4 EUR / month server.

That is horrifying... Don't sell me a virtual server if the hardware isn't remotely close to being available. I understand fronting me a bit on resources other than storage, but dang.

The annoying thing about that is when you create a server they ask money for reserved IPs (and if I remember correctly the HDDs) even if the server isn't being used or never spun up. I paid my 26 cents to them for nothing and those were the last cents they saw from me.

That happened to me too, they charged me for an IP address that I never got the opportunity to attach to a vServer (because there were no physical instances to spin the vServer on)

FWIW I have had a positive experience with their super low cost small ARMv8 instances, which help me test PostgreSQL patches-in-development on that architecture. (It'd be even better if they had a FreeBSD image available.)

On the "sell you a server but can't guarantee when you'll be able to spin up" part, they also don't charge you till it is running. It's not like they say: sing up, spend $x on a machine, but there's a 50/50 change it won't start, but Your charged anyway...

OVH/Soyoustart/Kimsufi (same company) at least doesn't let you pay money for, or order a server, unless it is "in stock".

A terrible experience for you, but our business model was validated as demand has been above capacity for years! Take that SI valley. La baleine échouée est maintenant la nôtre!

I checked them out after using GCP, AWS and its unbelievable how much slower their servers are. I some times wonder how anyone other than AWS/GCP/Azure even survive in this competitive market.

Many people think the same for GCP,AWS when you compare their VPS with dedicated hardware. Though the managed services are nice.

I’m using it for personal hosting and it’s way cheaper than others. The speed isn’t a big deal for a Nextcloud or Gogs server since I am the only user.

I wouldn’t use it for any large web service though.

If you only want a server, OVH/Hetzner are much cheaper for the same performance. To me, AWS/GCP/Azure only really make sense if you start using the whole ecosystem. Otherwise you can get the same for half the price..

I've spent the last 3 years doing AWS stuff at work and agree 100%. If all you're using is EC2 then it's not a great value. Where it's worthwhile are the other services you can tie in to really easily. At work we're using:

  SQS
  SNS
  RDS
  S3
  Athena/Glue
  EMR
  CloudWatch
  Lambda
  CloudWatch Events
  EC2 Autoscaling, ELB, ALB
  Elasticsearch Service
  Glacier
  ElastiCache
  Route 53
  CloudFront
  Data Pipeline

It's spendy, no question. None of this is stuff you can't do yourself on cheap instances. We'd just need an additional 2-3 FTEs to maintain the homegrown jungle of crap that would replace the managed services. Oops, there went my savings.

Price - traffic is included (while GCP/AWS charge a lot for it, and a sudden surge in traffic could lead to an unexpected bill), and the server costs a fraction of what a small server would cost on GCP/AWS (except for the shared-CPU instances which I found to be slower than Scaleway once the throttling hit).

If you want to run a business, Scaleway's lack of reliability (servers often sell out etc.) may be a problem. If you're a hobbyist, the potential bandwidth bill from GCP/AWS is a show-stopper, and the Scaleway servers are more than good enough for many applications.

Regarding the slowness, are you talking about CPU speed, I/O, or network?

Is one still required to submit all their personal information (ID card details, proof of address, etc.) before the servers can be useful? Previously you had to do it, or SMTP ports (etc.) would be blocked until you did.

If that's still the case, is it at least made known that the servers are crippled, prior to starting your account creation process (previously it wasn't, which I found shady)?

My experience so far is that the servers are okay, but the website's UI could be clunky-ish, support to be unresponsive and the aforementioned shady business with requesting personal-information to enable services.

so them blocking a port that can be used for sending Spam mail, without you identifying yourself, is shady?

Yes. But who cares, there are many better services.

Care to name the one(s) you recommend and why?

Rather not, I'm still glad it's not very well known in SV or mentioned on HN. Sorry.

I tried scaleway, and I'm glad I didn't try to move my personal SMTP server there.

The other issue was that I can't run my own kernel there, which was also non-obvious before purchasing. They make it look like it's baremetal, but it's limited in this way.

Now I'm on a fully virtualized vmware VM where I can do pretty much anything, including running a kernel with WireGuard included. And it's half the price of 1-XS for the same specs and no blocked ports and no uploading of my ID somewhere I don't trust.

Aruba Cloud?

Then you declare yourself wrong.

If I don't know better services exist then from my PoV they are not available and so may as well not exist.

While spam is a real issue and it's well justifiable for a company to request some identity confirmation, it is also not particularly wrong to label not disclosing the requirements upfront to be a sort of a dark pattern. It surely can be somewhat frustrating if you've signed up only to see that there are some extra requirements you haven't anticipated.

IMO, requesting identity is perfectly normal - albeit a bit annoying, but I get the point. Take Hetzner, for example.

Bullshit it's normal. Only a couple of hosts do this.

Don't try to normalize this stuff, there's no reason hosting companies should need scans of their customers IDs.

It's usually the low budget hosts that do this, since those will attract more shady "customers"

Easy way to get rid of those

Pretty much every physical hosting provider I use (i.e. a company that provides physical boxes or slots in a data center) have required a scan of an ID to setup your account.

So yeah, pretty normal.

If you're physically going into the DC, yeah that's normal.

ID scans for colo with remote hands? Nah.

We're quite far from either of those anyway.

I don’t believe they do this now.

I am using Scaleway for hosting my own mail server and never sent them an ID. They have my credit card info, of course, but that’s it.

How are you running your email? Is it you primary email?

What's the catch with the unlimited traffic? Can I use their servers to deliver large content files to a large group of users, with only the advertised bandwidth as the limitation? Or do they decrease bandwidth after a certain amount of traffic?

I ran 173TB combined up/down last month without ever being throttled on their C2L instance. I've run 583TB through it in the last 5 months. It even bursts to 2.2Gbps every few minutes.

What did that cost you? What's the bandwidth to other countries like?

Scaleway doesn't charge for bandwidth, so I'd guess their monthly cost is the price of the C2L (23,99€/month)

Been running high-bandwidth stuff there for a while, no issues. They really don't seem to mind high-bandwidth applications.

I guess for every person that uses 100 TB a month, there's the person who runs an IRC bouncer. I know that at least my Scaleway instance only has one single-user Django app, a static website and an IRC bouncer, so the traffic amounts must be tiny.

Last I looked at Scaleway what put me off was the fact that their storage was not backed by RAID [1].

Compared to other, comparably priced providers like Hetzner and OVH that do use RAID this is a big caveat IMHO. Of course important data should be backed up remotely but disks fail and I'm not willing to deal with the hassle of restoring data simply because a single disk failed.

[1] https://www.scaleway.com/faq/servers/volumes/#-Local-Volumes

Different use cases. If your servers are sufficiently redundant, then RAID is a waste of money, and you might be better off getting more capacity for your money.

But for critical data, or data where you can't scale out as easily, of course, that's another matter.

I won't use RAID for services I have 10 identical copies of and health checks, but I will for the database server that is a pain if I have to fail over or restore, in other words.

From my experience Scaleway is awfully slow and you can't really scale your project as you like as there is often no stock. There is also that weird IP thing where you need to bind to private IP instead of the public. This causes a lot of problems - for example using custom dns for Kubernetes doesn't really work with that setup. Only thing good about it is the network. Another thing - if your instance has a fault and there is no stock you'll have downtime, sometimes even a day or so until there is something in stock.

Personally, I don't think Scaleway servers are good enough to be used in production like you describe it. However, because of the low prices, they are ideal (for me at least) for hosting small web apps.

That is sadly true. I am actually going to delete my cluster tomorrow because of that.

They are known for overselling a bit. They admit it in this very blog post. Nice technical, leading edge concepts, good prices. But then they might struggle with fulfilling the promises.

They had big announcements for IPv6. After a long wait they had to admit that their C1 infra has HW limitations and IPv6 would never come. On their other infra they do have IPv6, but I understand the implementation is awkward (not an IPv6 expert myself).

They had big plans for ARM. Nowadays they do mostly Intel.

Yes, I like them for pioneering, but I would not like to have my business heavily relying on their promises.

Forgot to mention: You get free networking on IPv6. Certainly a good deal, if you have usage for servers without IPv4.

Does someone have a benchmark of the new cpus in the start-instances vs the older X86 baremetal servers, C2S and higher?

Not baremetal, but I did some benchmarks before and after the free upgrade of my VC1 to Start1 hardware.

With UnixBench single thread:

--------------------------------------

Old VC1: Intel Atom C2750 @ 2.40GHz

Dhrystone 2: 12029634 lps

Double-Precision Whetstone: 1984 MWIPS

--------------------------------------

New Start1: Intel Atom C3955 @ 2.10GHz

Dhrystone 2: 22143815 lps

Double-Precision Whetstone: 3165 MWIPS

--------------------------------------

Other tests that I performed also showed a ~60% performance increase, except for some compression tests that only gained ~45%.

I'd love if Online.net followed OVH's footsteps and opened up a North American DC (Canadian would be great).

I'd jump ship to OVH if they have a DC in NY or SF.

What's nice is that simply turning the server on and off allows you to move the faster machines :-)

Except that you may suddenly encounter that servers were out of stock, someone just got lucky with the slot you've freed and now you can't power your server back on. ;-)

Not sure about VPSes, but this had happened to me once with a C1 server.

Did it happen to you on a VC1 (now Start-XXX) ? These are virtual servers.

No, I've only ever used their baremetal offerings (C1 and C2s). Sans the few oddities, I think they're quite good for the price, at least for various toy/sandbox stuff. It could be that this kind of situation can't happen on VC1 (Start-XXX) - as I wrote, I'm not sure.

More details are needed, it looks like a good story.

Well, I don't exactly remember what I was doing (it was, like, an year ago or so), but I've de-provisioned a C1 to temporarily release its system volume as I wanted to attach it to another server (to make changes there). IIRC, I've messed up my root partition a little bit, had some nbd issues with recovery kernel, and decided to just reattach the volume to another C1 instead. And then, basically, what I wrote - someone got lucky and got the computing resources I've freed, so I wasn't able to spin things up.

How?

See the end of the blog post:

> All VC1 users can upgrade to Start servers without any price change.

> To upgrade your existing VC1 server, simply do a power off and a power on your server. Your server will start on a Next-Gen Start Hypervisor with increased performances and for the same price as before. You may already be running on a new hypervisor if you created your server recently. New hypervisors are powered by Intel Denverton CPUs instead of Avoton.

I tried doing this to my AMS1 region VC1L server but after powering off/on the server it was still using the old hypervisor (C2xxx atom). I noticed that the Start Range servers are all out of stock so I suppose they did fallback to old ones.

(I've already opened a ticket with them).

Edit: Just received a second message from them. Turns out I a missed part of their blog post that says that the update is not yet available in AMS1 region.

Reply on news.ycombinator.com