New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
How good is Scaleway?
elwebmaster
Member
in General
Care to share experiences with the Scaleway VMs (not the ARM servers)? I am looking to use them in production, we are talking about 100 VMs of which may be 10 will be ON at any given time.
I tested them and performance appears good for my needs so far. I need reasonable I/O, latency and CPU, which the AMS instance I used for testing delivers. But has there been any major downtime in the past or other issues?
Comments
Link between FR and AMS is on 104% load since days, no one seems to give a shit.
Even the Uplinks to google are full.
Did you make any network benchmarks?
I m using it for a production site since last 8 months with no issues. Only draw back I see is to take a snapshot we need to shutdown the server.
And that removes your monthly charging cap for that server, so the 2.99 EUR one will cost 4.46 EUR/month, and that gives you a new IPv6 (since those are dynamic) after the restart, and most likely places you on a different node entirely, with a different amounts of CPU oversell (may end up worse).
You really don't want to be stopping and starting Scaleway, it's 1.5x more expensive if you do.
For my use case I think that's fine. There aren't too many options with hourly billing and 2GB RAM, not only at 4.46 EUR/month but even at 10EUR/month. Most providers (including OVH Cloud) bill you for all resources even when the VM is OFF. The only other feasible options I found are LunaNode and iwStack. So my plan is to start with Scaleway, and may be add these two to the mix as a backup if things start going south with Scaleway.
OVH VPS-SSD at the same 2.99 EUR is a better option at this point.
It does have the same hourly/monthly billing difference in price (when ordered as "cloud", not as "VPS"), but considering the other Scaleway issues such as overloaded network, dynamic IPv6 and more CPU oversell with worse CPUs in the first place, it still wins.
Problem is not charging when it's off, but if you simply turn it off for a minute mid-month, all the usage you had before and after that, now costs 1.5x more.
Why hourly billing is a requirement for production server ?
Traffic spikes or sneakers
OVH charges when the Cloud VPS is OFF, they only stop charging when you "terminate" (as in "delete") the server, at least that's what I understood from their website. In this case for my application I will need to snapshot the server, terminate it, then when it needs to go back up restore from snapshot. How long will that take for a 10GB server? I guess I need to test it, but it won't be acceptable if takes more than 2 minutes to restore from snapshot.
It's production as in I will be selling a service built on them, but the service itself is sandbox-like, so most customers will probably use it sporadically / periodically, not 24/7.
No.
The file systems have no RAID, which immediately makes them shaky for any production service. Maybe you can mitigate that with database replication etc., but then spinning hourly instances up and down gets even messier.
Perhaps you should instead get a big dedi and run proxmox on it to self-host your ephemeral VM's.
Good point about the lack of RAID. This is something we need to communicate to our customers. But Scalway do say that volumes are archived to permanent storage on start/stop so I hope that means the risk of data loss is only while the VM is on.
We got dedis and configured them for VMs but then we realized the risk and time commitment to manage them is not worth it. Our focus as a business is on what runs inside the VM and managing infrastructure is not something we can afford at this point.
If you're running a service business of the type you describe, I think you have to actually solve the problem and not just inform people of it. They will expect their data to not be lost no matter what you tell them ahead of time.
I had very mixed experiences with their service and was not really happy in the end. Lots of CPU steal and wonky network from time to time. There are much better solutions out there at a comparable price point. (OVH being one of them)
I hear that more about AMS than Paris fwiw.
So basically trash that only hipsters use?
I wouldn't say that. I use it for various things and like it. Just not for production services.
I tested it a bit. Didn't take screenshots of benchmarks but it mostly went like this :
Took the 8 core 8 gb ram package
Took 2 hours to install cPanel, after installation, server loads at 2.0 right off the bat ( with nothing else running )
Told me I needed many documents to lift the SMTP port block that they enforce.
Is very hard to mount additional disk drives properly.
Closed the server, got billed a like 0.04 euros
Great!
The CPU is just too slow.
CPU is just too slow.not supported Windows System. Network not good for Asia.
Do you get better connection to OVH BHS since that's the other option (LunaNode)? Also, is it high latency or low bandwidth that's the issue?
Too bad to be used in production enviroment?
Can't really comment for production scenarios but the one plus it has over OVH is the ability to exceed 100Mbit. Speeds do periodically go to shit.
Also once something funky happened with the network configuration / virtual network card or w/e. I got some corrupted packets on any transfers that only got fixed after I archived the VM and re-deployed. Restarts and nuking configs to let it recreate it them didn't help.
I hadn't touched the VM in months but it might have been something more easily corrected by someone more experienced than me.
We did some benchmarks, before they upgraded the cores:
https://wiki.x8e.net/doku.php?id=dedicated_benchmarks
Its basically a Atom or worse.
Question: unlike Scaleway, on OVH cloud is there any option to reserve IPv6?
I've used Scaleway for encoding 24/7, burning CPU's and never had problems with them, I had installed Windows, there was no downtime or anything. However they are slow (CPU).
So, I did benchmarks on those with more Cores, they said:
"Basically, if your application is able tu use multiple cores, you will get almost a 100% performance increase"
ARM 2gig, before: 609, now 936, 65% increase.
ARM 4gig, before: 948, now 1293, 73% increase.
Source: https://wiki.x8e.net/doku.php?id=dedicated_benchmarks
For me 65% and 73% are not "almost a 100% performance increase".
But for 2.99, about 1k CPU Performance not dedicated is not that bad.
You could also go buy a Kimsufi for 5EUR and get 0.8k dedicated.
Looks like Scaleway is down?
Looks like Scaleway is down?
FS#936 - [Scaleway][PAR1] Network issue
We've detect an issue on one of our router in PAR1 and nodes is not reachable
Issue is under active work of our team, we do our best to solve it as soon as possible
Thanks for your understanding and patience
@elwebmaster Only use Scaleway bare metal, the rest are virtual machines.