[TRIAL] BuyVM needs people to help test vswap

Jeffrey

OK, in WHMCS it says the status of my VPS is active, but I cannot get in the control panel and I did not receive any emails regarding the login for the server.

giaythuytinh176

My order number: 6732142761
Please active my VPS.

yomero

@Francisco said: even with the node sitting at 400 loads from people forkbombing prime95 >_>

LOOOL WTF, that is brutal abuse u_u

miTgiB

@yomero said: LOOOL WTF, that is brutal abuse

Naw, I had loads over 1200 from fork bombs during my beta

yomero

HAHAHA, that's nasty xD

I am just loading a bit now.... :P

vedran

Crashed again? Or you guys are just terribly abusing the node?

Ash_Hawkridge

@Jeffrey

Calm down im sure they will get you online soon, its not like your a paying customer :P

Ash_Hawkridge

Got me a vSwap VPS :P Your security questions are some of the most trivial i have seen haha.

Francisco

@VMPort said: Got me a vSwap VPS :P Your security questions are some of the most trivial i have seen haha.

vswap is actually a major problem and the cause of all of the crashes it seems. We've modified stallion to actually just give users 'burst RAM' without any vswap and all of the crashes have gone away, even when users are memtesting it

It's stable and runs good, i'm happy so far.

Francisco

Francisco

@yomero - i'm noticing some Q3 servers running, is that you? How is it running? is the RAM allocating properly?

Francisco

Ash_Hawkridge

@Francisco

So the goal is to use RHEL6, but without the most important feature that RHEL6 brought to the table for OVZ :P

Francisco

@VMPort said: So the goal is to use RHEL6, but without the most important feature that RHEL6 brought to the table for OVZ :P

We still get the benefits of the new memory management, just there's no vswap in place. If anything, this helps the clients a bit more since their stuff won't degrade performance wise :P

For what it's worth we're using an RHEL5 userland (centos 5) with the .32 and then a nice coat of our own adjustments/improvements to hopefully address the stability issues.

I'm seeing that node99 has been stable since we applied the latest round. I'm seeing node98 was rebooted at one point but I'm waiting on Anthony to wake up to confirm as to why (panic or if something was soft locked)

Francisco

Ash_Hawkridge

@Francisco

Right im with you. Its nice to see somebody actually trying to work around the issues instead of just ignoring them and rolling back to 5.7. I hope it all works out for you

I finally put in my order as well

Francisco

@VMPort said: Right im with you. Its nice to see somebody actually trying to work around the issues instead of just ignoring them and rolling back to 5.7. I hope it all works out for you

By using an RHEL5 userland we can actually roll back to a .18 kernel without many issues Granted, we're hoping that all of the issues are because of vswap and not bigger issues in the source.

node99 is running the latest RHEL6 testing kernel, where as node98 is running a proxmox 2.6.32 kernel from October, probably before any of the big updates have come in I'll probably roll 98 up to the same as 99 to see how it goes.

Francisco

Francisco

@VMPort check this out

/usr/local/crashplan/jre/bin/java -Dfile.encoding=UTF-8 -Dapp=CrashPlanService -DappBaseName=CrashPlan -Xms20m -Xmx512m -Djava.net.preferIPv4Stack=true -Dsun.net.inetaddr.ttl=300 -Dnetworkaddress.cache.ttl=300 -Dsun.net.inetaddr.negative.ttl=0 -Dnetworkaddress.cache.negative.ttl=0 -classpath /usr/local/crashplan/lib/com.backup42.desktop.jar:/usr/local/crashplan/lang com.backup42.service.CPService

Which turns into this:

(05:05:12) node99:~ root: vzctl exec 14990 free -m
             total       used       free     shared    buffers     cached
Mem:          1024         71        952          0          0         23
-/+ buffers/cache:         48        975
Swap:            0          0          0

If you check the Xms is being abided by, even without swap being defined. Another interesting thing i've been discussing with @mitgib on IRC is that:

• node99 had 3GB RAM of total usage
• node99 has 24G RAM in total
• vm.swappiness=0 meaning it should only swap if we're at 23.5GB of RAM usage or so
• node99 was somehow 700MB into swap??
• once vswap was disabled on all CT's the swap usage went down to 0KB

This was with vswap enabled on CT's, @mitgib actually documented having the same issue during his vswap trials.

Francisco

Taylor

Mines marked as active but no information revived about it /:

yomero

@Francisco said: @yomero - i'm noticing some Q3 servers running, is that you? How is it running? is the RAM allocating properly?

Francisco

OMG you catched me abusing! u_u LOL. Yes I am

Well, yes, the RAM have the normal behavior of a Xen/KVM/real machine n_n Still worried about the CPU monitoring, but I guess that "context switching" and probably innacurate loads is part of OpenVZ and can't be fixed.

So far stable (14 hours!) And lagging hard as it must be (in-game lol)

yomero

@Francisco said: If you check the Xms is being abided by, even without swap being defined. Another interesting thing i've been discussing with @mitgib on IRC is that:

node99 had 3GB RAM of total usage

node99 has 24G RAM in total
vm.swappiness=0 meaning it should only swap if we're at 23.5GB of RAM usage or so
node99 was somehow 700MB into swap??
once vswap was disabled on all CT's the swap usage went down to 0KB

Sorry double post.

To this I must add an example. Have you used Google Chrome in a low RAM machine?

Well, in this situation I am very confused, I have a netbook with 1.5GB RAM. Despite using about 500 MB (according to htop), the netbook starts to swap like hell with only a few tabs. And no, isn't because I have a bad value in /proc.../swappiness. I guess is the WEIRD way that Chrome uses to manage memory. Maybe other software have the similar problem. (Java?)

HerrMaulwurf

Pls refill stock

Jeffrey

Same here @Taylor. I guess we just need more patience

ztec

Power to the people.

dmmcintyre3

I have been having memory allocations fail while only using 300mb of 1GB RAM.

yomero

@dmmcintyre3 said: I have been having memory allocations fail while only using 300mb of 1GB RAM.

I can use all the 1GB RAM o_O What are you trying to use?

dmmcintyre3

java/minecraft

yomero

Maybe the no swap is affecting you?

dmmcintyre3

http://d3vm.net/bvm-vswap-beta.txt

privvmpages is set to 2gb and almost out but only 225-250mb is reported as used by free -m

yomero

Huh, I am afraid that is for Java? :P

Francisco

@yomero said: OMG you catched me abusing! u_u LOL. Yes I am

Well, yes, the RAM have the normal behavior of a Xen/KVM/real machine n_n Still worried about the CPU monitoring, but I guess that "context switching" and probably innacurate loads is part of OpenVZ and can't be fixed.

So far stable (14 hours!) And lagging hard as it must be (in-game lol)

The node is getting raped, every bit of CPU is getting used up so the stats are accurate :P

This is load testing, not what it'll be like if/when we roll it.

Francisco

Infinity

Yay, I managed to pick one up via my really crappy phone.

https://www.youtube.com/watch?v= HuYHxAKRpo4 -- a video of my phone made for a forum's "screenshot your phone thread".

Francisco

@dmmcintyre3 said: privvmpages is set to 2gb and almost out but only 225-250mb is reported as used by free -m

It's likely screwing up because your physpages is set to only 1GB.

physpages is obeyed first, then the others.

Francisco