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.
OpenVZ Burst/Swap
It seems that burst is a better marketing term for OpenVZ and that naming what OpenVZ does swap, does it no justice.
Are you guys that are running the latest OpenVZ still referring to the ram as Burst? The limitation that was put on 'Burst' was that it emulates slowing down the container so now they call it swap.(but does it really?) It's not using real swap either, it's using RAM just like the old burst.
Comments
my impression is memory accounting is better when there is swap and not burst? hence preferable for customers? not sure, i haven't signed up with someone with swap.
@jcaleb well when you advertise your product you can say
512MB Gauranteed 1024 burstable and they think they will get 1024MB ram
but when you say 512MB Gauranteed 1024MB Swap they think they are getting 1512mb of ram and more than half of it is 'slow swap'
OpenVZ with vSwap has much better memory management since the memory limits are not allocation based anymore.
@dmmcintyre2 it does and I have noticed that on our nodes with vswap. - but like I said about the marketing thing ... they sure do try to make it harder
vswap ...it's value, on a high performance system, cannot be lightly dismissed. It's been the subject of a lot of reading for me lately.
512MB Gauranteed 1024 burstable and they think they will get 1024MB ram
but when you say 512MB Gauranteed 1024MB Swap they think they are getting 1512mb of ram and more than half of it is 'slow swap'
some providers advertise as 512Mb Guaranteed, 1024Mb burst (512mb swap upon request). that way you hit best of both worlds
@corey-} does vswap actually takes a stab out of the Nodes Ram or it simply emulates swap without any physical hassel or something else I am using. This is the first time I am going to provide vswap instead of burst ram so I should better do my calculation.
While vswap is a nice feature and one that we would like to implement, I still do not find CentOS 6 mature enough to utilize it and to run OpenVZ on it. In my lab, the tests with CentOS 6 have not been pretty =(
It is "slow swap", although only virtually, as I understand it...
vSwap is artificially slowed down and if the node runs oom it is the first thing to be sent to the swap.
CentOS isn't really the issue, it's just OVZ's inability to fix their bugs. They claimed they fixed a major crash bug but it wasn't handled at all:
http://bugzilla.openvz.org/show_bug.cgi?id=2278
Using the kernel is simply a bad idea, @Daniel can attest to that.
Francisco
Exactly.
In the past you ignored me about this, maybe this time I can be more lucky :-)
Maybe there are some specific cpu issue? When you warned me about the problems with the .32 kernel I had in the logs some softlock on cpu with some high load spikes in the graphs and one crash (on one out of 3 nodes). Since the 042stab053.5 no more softlock were logged and no more spikes on graph.
Now I know you had one node crash after >80 days, so I'm waiting if the node can pass that mark point...
My nodes run on AMD Opteron 6176 or on not so new Intel cpu with no HT
Lately we bought 2 supermicro microcloud with E3 and I used one of the 16 servers to play with openvz and bang! using the same kernel I was experiencing locks and panic...
I didn't ignore you, I just know it isn't that
I tested with an E3, an L5520, L5420 as well as have friends with AMD opterons, all crashy.
Now, we had 2 nodes (node55 & 56 I think?) that were 100+ days uptime on a 49.6 based kernel. No idea how that was possible, but it was without issue.
The latest OVZ's have been a mess where as a few in the 4x's were "ok".
By all means, check their bug tracker and you'll see gobs of related crashes. It took them 6 months to address a crash I was reporting back in December when we were doing our public vswap tests.
Hell, most hosts out there use our "don't enable vswap" trick in hopes of stabilizing things.
Francisco
Ok, then I must knock on wood for one more month at least :-)
This is the very first reason why I went straight to KVM with the new hardware... Lack of confidence/trust.
Thanks :-)
We've not had one crash in 6 months of running VZ/Vswap/CentOS6. Although now I've said that, I'm sure one of them will burn
We haven't had any crashes either.
do all of you mean vswap is not mature and we mortal users should avoid it for now?
@jcaleb no someone is claiming with their configuration a certain version of the openvz kernel is crashing.
Swap should be avoided in any case. Optimize or get better spec
@NinjaHawk Desktop OS's load everything into ram without swap, then don't use the swap if it's there, stops loading everything into memory, and performs better. Don't ask me why, no one seems to want to touch that question, but it's certainly a reason to have swap, if you like to get the most out of your memory without building the entire thing up package by package and rewriting configs for hours, just for a GUI desktop.
@NinjaHawk noone here is using swap.
It uses ram.
Ninjahawk is a prime example of why vswap is a bad marketing term.
@Corey I like Ubuntu desktop. Can't use it with under 950mb ram usage, or swap being there. Safe to say I use it. I fake swap in burst setups otherwise and get the result I desire...until I find an app that tries to use it.
@Corey what did I do?
@jarland I meant those who completely relies on additional swap. Once in a while ok but if this is a regular thing, something else is wrong.
You said "Swap should be avoided in any case. Optimize or get better spec ", but noone here said they were using swap.
I did not say that. I meant ... well wateva.
I just meant is general. If swap is something that is happening on a webserver, time to investigate a bit more?
Well just saying.
@ninjahawk so as a consumer you would prefer burst ram right? Because it doesn't have the name swap in it?
I don't understand how it can be just a marketing term. Swap and burst are not the same so long as systems operate differently with swap than without it.
@jarland has anyone done research to see just how much this vswap 'virtually' slows down the container? I doubt it's as much as real swap.
@Corey as a consumer, I am happy with my guaranteed 128 ram I do not need 1024 swap or burst.