[CentOS-virt] Virtuozzo GFS
I have just finished deploying two Dell PowerEdge 1950s with CentOS 5.1 and Virtuozzo 4. GFS is up and running and Virtuozzo is configured for shared-storage clustering. Everything works adequately but I am wondering if anyone else has experienced load issues like I am seeing. I have three VEs/VMs running, two on one node and one on the other node. One of the VEs on each node are doing very little (one is just idling with apache and mysql and the other is running rsync every six hours). The other is running Zimbra. Every so often load will spike on the node running the Zimbra VE to as high as 2 or 3 then settle down a short while later to around 0.8 or 0.9. During the spikes the node not running Zimbra will other see an increase from its idle load of 0.4 or so up to as high as 1.7 as I have seen. I notice when running top that dlm_send and dlm_recv will jump to the top fairly frequently when these load spikes occur. What I am wondering is whether anyone else has experienced these kind of load scenarios with GFS and what they have done to deal with them? We are hoping to deploy a bit more densely on this setup so I'd like to make any performance adjustments I can at this stage. Thanks, James Thompson ___ CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
Re: [CentOS-virt] Virtuozzo GFS
James, - James Thompson [EMAIL PROTECTED] wrote: I have just finished deploying two Dell PowerEdge 1950s with CentOS 5.1 and Virtuozzo 4. GFS is up and running and Virtuozzo is configured for shared-storage clustering. Everything works adequately but I am wondering if anyone else has experienced load issues like I am seeing. I have three VEs/VMs running, two on one node and one on the other node. One of the VEs on each node are doing very little (one is just idling with apache and mysql and the other is running rsync every six hours). The other is running Zimbra. Every so often load will spike on the node running the Zimbra VE to as high as 2 or 3 then settle down a short while later to around 0.8 or 0.9. During the spikes the node not running Zimbra will other see an increase from its idle load of 0.4 or so up to as high as 1.7 as I have seen. I notice when running top that dlm_send and dlm_recv will jump to the top fairly frequently when these load spikes occur. What I am wondering is whether anyone else has experienced these kind of load scenarios with GFS and what they have done to deal with them? We are hoping to deploy a bit more densely on this setup so I'd like to make any performance adjustments I can at this stage. Thanks, James Thompson A load of 2-3 isn't much at all... so I don't think I'd call that much of a spike. I have run OpenVZ at work and on a hobby server. In both cases I have about 7 containers... one of them being Zimbra. The other 6 containers are fairly busy so the two machines see a decent amount of load. I am NOT using GFS though. What is dlm_send and dlm_recv part of? GFS? TYL, -- Scott Dowdle 704 Church Street Belgrade, MT 59714 (406)388-0827 [home] (406)994-3931 [work] ___ CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
Re: [CentOS-virt] Virtuozzo GFS
I believe both dlm_send and dlm_receive are part of GFS' locking mechanism where locks are bouncing back and forth between servers. The Zimbra VE is pretty consistent in its internal load and what struck me was that Virtuozzo's PIM interface flags the increased load (leading me to believe it is abnormal). Perhaps I am mistaken on the meaning of the load figures though. On a busy machine what would be considered a normal load? Does it differ based on the number of processors and cores available in the system? I came across this article which makes me think I may be mistaken in my concern: http://www.linuxjournal.com/article/9001 Can anyone who is using GFS perhaps comment on how it has performed for them in virtualization scenarios? When initially working with just the iSCSI mounted disks involved operations were just slightly less snappy than local file operations. With GFS things seem at least noticeably more sluggish which I expect is normal given the multi-server locking but has anyone worked on optimizing GFS performance in a virtualization environment? Thanks, James Thompson On Sat, May 31, 2008 at 11:56 AM, Scott Dowdle [EMAIL PROTECTED] wrote: James, - James Thompson [EMAIL PROTECTED] wrote: I have just finished deploying two Dell PowerEdge 1950s with CentOS 5.1 and Virtuozzo 4. GFS is up and running and Virtuozzo is configured for shared-storage clustering. Everything works adequately but I am wondering if anyone else has experienced load issues like I am seeing. I have three VEs/VMs running, two on one node and one on the other node. One of the VEs on each node are doing very little (one is just idling with apache and mysql and the other is running rsync every six hours). The other is running Zimbra. Every so often load will spike on the node running the Zimbra VE to as high as 2 or 3 then settle down a short while later to around 0.8 or 0.9. During the spikes the node not running Zimbra will other see an increase from its idle load of 0.4 or so up to as high as 1.7 as I have seen. I notice when running top that dlm_send and dlm_recv will jump to the top fairly frequently when these load spikes occur. What I am wondering is whether anyone else has experienced these kind of load scenarios with GFS and what they have done to deal with them? We are hoping to deploy a bit more densely on this setup so I'd like to make any performance adjustments I can at this stage. Thanks, James Thompson A load of 2-3 isn't much at all... so I don't think I'd call that much of a spike. I have run OpenVZ at work and on a hobby server. In both cases I have about 7 containers... one of them being Zimbra. The other 6 containers are fairly busy so the two machines see a decent amount of load. I am NOT using GFS though. What is dlm_send and dlm_recv part of? GFS? TYL, -- Scott Dowdle 704 Church Street Belgrade, MT 59714 (406)388-0827 [home] (406)994-3931 [work] ___ CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt ___ CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
Re: [CentOS-virt] Virtuozzo GFS
On Sat, 2008-05-31 at 16:19 -0400, James Thompson wrote: With GFS things seem at least noticeably more sluggish which I expect is normal given the multi-se James, I haven't personally used GFS before, but the sysstat package might be able to help you out. Basically, from what i can see there would be two different causes of the load: CPU: We've ran Zimbra in VMWare Server with an average load average of around 2-3 (exactly what your seeing) using Full Virtualisation via Intel VT (ensure your Dell 1950 has this turned on via BIOS - it's not on by default). Here you would probably see 'top' claiming the load average (on the hypervisor) to be mostly on CPU. IO: Use the sysstat rpm (iowait app) to have a look at iowait times and narrow it down to a IO problem, here you'd see 'top' claiming high load average but CPU's look fine. We ran iSCSI on a 50mbit link (quick test), and whilst formatting the load average shot up to ~5, CPU and everything was fine and iowait quickly showed us it was waiting on IO the whole time. Hope that might help, you've probably checked these already though. Does Virtualizzo supper paravirtualised guests - that might be the next step to check out if you haven't already... ___ CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
