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.
Why don't OpenVZ providers go to vSwap?
Why don't OpenVZ providers go to vSwap?
I have OpenVZ accounts with burst, and OpenVZ accounts with vSwap.
On a clean re-image and install of 20 new WordPress sites via LEB nginx/wordpress script, a 128/128vSwap host uses ~40 megs less than the same install on a 128/256burstable OpenVZ host.
vSwap FTW.
OpenVZ providers... PLZ, Pleeaaassseee go to vSwap.
Thank you.
Carry on.
Comments
We have been on vSwap since it was supported.
We are using vSwap. Do not see a reason why with vSwap the memory usage will differ with the same started processes ...
Me either.
It's mostly because of FUD (Fear, Uncertainty and Doubt) - about the 2.6.32 (aka vswap capable) OpenVZ kernels. There were a lot of bugs / problems with these kernels in the past. I guess most of those bugs should be fixed by now, but there is still the FUD factor.
Applications often take up more memory if swap is not available. Especially true of the Gnome desktop environment that I so love. Can be 150-200mb with vSwap or well over 1GB without. You can witness it by creating a fake swap file on a non-vSwap OpenVZ vps and understand how nearly irrelevant the swap is to most actual usage (obviously a relative statement, thus the "most" word), but the mere presence or illusion of swap changes how applications load themselves into memory.
Swap is vital to creating an environment that feels more like true virtualization. Typically packages are not designed with the purpose of being placed on a system with no swap available.
That said, it may be unstable in some systems. I don't have an issue with it on nodes with 60 people, but I can't speak for someone who has higher capacity systems and more usage.
How is this accomplished?
Got a link?
Make this into a bash script and run it. It's fake. You'll probably crash if something actually tries to use the swap. I've loaded the entire Unity 2D desktop without it ever trying to use this fake swap, and lowered my memory usage by 1/4.
Probably doesn't have an effect on already started processes.
The instability of vswap (rewrite of OVZ) compared to the burst only ones is real. Much progress has been made, however, the only crashes we experienced at prometeus were on nodes with OVZ, sometimes impossible to explain, see the infamous node 22 crashes from last summer and one even on node 24.
This is why I said in an uptime thread that ovz should generally not be compared with other real virtualizations, because of the instability problems. It is also not fair because some providers offer burst still and that cant really be compared with vswap regarding features and usability, but will have a much better uptime. More or less stable vswap kernels came out last summer only.
I honestly think the difference between those of us who experience no crashes with vSwap and those who do is node capacity. It may cripple under a higher load, which isn't a statement of "oh you oversell more than me" ...more likely you'd have more cores and memory than me and therefore a higher capacity and more usage.
I have EL6 nodes, just choose not to use vswap, many people have reported stability issues.
john@server1:~$ free -m
total used free shared buffers cached
Mem: 1536 3 1532 0 0 0
-/+ buffers/cache: 3 1532
Swap: 512 0 512
Ermahgerd! It is totally misreporting memory usage.
I currently have a full Debian LAMP stack with a large Joomla site and a few smaller sites running, and am logged into XFCE via nxserver.
3 megs? LOL
Response is very snappy. Firefox loads in 10 seconds instead of 30.
Also, FreeMind is actually usable.
We have been using it on our OpenVZ line. Its been very stable for us.
Is getkvm.com down right now? I followed the link from your sig, it timed out. Tried a proxy, it still timed out.
Also, http://www.downforeveryoneorjustme.com/getkvm.com
No one said it yet, but I guess the biggest reason for why they may not offer it 'yet', is because they would have to upgrade their CentOS 5.8 boxes to 6 in order to do so (or only start offering them on new boxes).
Far as why non-vSwap uses more memory, is because in the previous configuration OpenVz allocates more memory per process to avoid OOM errors, a single nginx process could take up to 50% to 100% more memory than the equivalent on OpenVz+vSwap, KVM or Xen. Thus why you actually need a higher memory ceiling. Someone else had a better explanation of it on LET months and months ago.
That does make sense. Uptime FTW!
We have 3 OpenVZ nodes with CentOS 6 using vSwap with 100+ days uptime and no stability issues. Total node RAM was never used more than 40-50%. May be when the RAM is ending (if oversold) start having some problems.
@qhoster I think you mis-interpreted the "Uptime FTW", I was saying some may not have it yet because would require taking down boxes with CentOS 5.8, backing up all the VPSes on that node, installing CentOS6 and reinstalling all the VPS users, or simply create a new box with CentOS6 on it and start adding users to that one with the vSwap option.
Also, http://www.downforeveryoneorjustme.com/getkvm.com
Yup, under a traffic flood at the moment
Our email is still up however if you need anything, [email protected].
Damn skids.
Yes, only the big nodes crashed.
And, no, it is not an oversold or not question (tho it might have something to do with the frequency of crashes), at the time node 24 crashed was at only 30% capacity IIRC.
That changed since last summer, from early september kernels that didnt happen anymore (I mean lockups and the like). Touch wood, hope it doesnt happen now
2.6.32, all running a couple hundred containers with vSwap:
There are more, but I got tired of logging in and copy/pasting. Still waiting for instability apocalypse to occur..
@Damian how much ram on those servers ?
Here only 2-3 servers crashed in the last 6 months and all those were large OVZ nodes.
@Maounique: all of them are 32gb, except for that one listed at 105 days which is 24gb. They all tend to stay at around 40-60% RAM-utilized.
Here 2 64 and one 128, so jarland has a point Uncle also observed. Larger nodes with OVZ are more unstable.
So you basically way your options, take up less rack space and conserve money with more people on 1 node or choose stability with less customers and more cost.
Welcome to the game
I chose not to go vswap. And right now we are happy with it. Maybe in the future we will reconsider but right now we are perfectly happy.
Do you work there now?
All our new nodes are RHEL6 with swap support. However, the old once are still running with Burst memory. There is no difference in the stability, atleast on my side.
For some time. Made it in paper too starting last month.
Our 32GB OVZ w/vSwap nodes run fine, haven't had a crash yet