Not sure if this is the correct forum to talk about virtual hardware, but here goes:
We currently run Moodle on a small server running SUSE Enterprise 9. This server is also a virtual machine (using VMware server) on a SUSE Enterprise 9 host. We have had a few lockups recently and had to restart the virtual machine. Because of these lockups and because of the rapid growth in the number of users using Moodle, as of late, we are looking at new hardware. We are also considering a more enterprise class VMWare called ESX. Has anybody had any problem using VMWare with Moodle and SUSE?
Thanks,
Tim Nelson
Systems Administrator
Augustana College
For me the question is, if you hit the limits of VMware, why you put up the biggest VMware on the biggest machine you can get?
May be just stipping the present host would do the trick?
May be just stipping the present host would do the trick?
We are not going to put it on the biggest machine we can get. I just said we were putting it on new hardware. It is currently running on an old PC server, as it was originally setup that was for testing/beta. As we have allow more professors to use Moodle, we have out grown our old machine and are starting to have problems using VMWare. I do not know for sure if we had "hit the limits" as you mentioned. However, we need to make some changes to our hardware soon.
We are starting to use VMWare throughout our IT area, as it gives us more failover options and makes a server more portable between hardware, etc.
Before we make the jump to VMWare ESX (which we are likely to do on Moodle, as well as other servers) I want to know if any body here has had experience with it and if so, what issues arise with Moodle.
We are starting to use VMWare throughout our IT area, as it gives us more failover options and makes a server more portable between hardware, etc.
Before we make the jump to VMWare ESX (which we are likely to do on Moodle, as well as other servers) I want to know if any body here has had experience with it and if so, what issues arise with Moodle.
Hi Tim,
Interesting question. It's one I've just spent quite a bit of time recently trying to answer for myself.
My environment:
IBM Bladecenter model 8677
IBM Bladeserver LS20 dual CPU dual core 2 GHz Opteron (4 cores)
8GB RAM
200 MB/s fiber channel SAN
Nexsan array, 512MB cache, 8x500GB 7200RPM SATA drives in a RAID5 pack
VMware ESX 3.0 Enterprise
We have two of these blades and run about 14 VMs across the two, with a massive amount of CPU cycles to spare. We rarely run more than 10-15% CPU utilization across both the blades. Memory usage runs about 4-5GB per blade.
We've been running an ESX VM of SuSE Linux Enterprise Server 10, Moodle 1.7, mysql 5.x, Apache2, and php5. I had been using the uniprocessor 2.6.27.x kernel and only 1 CPU core. One of our faculty had his class using the chat server on Friday. Unfortunately I wasn't involved at the time, but I'm told the system hit the wall prompting a less experienced member of our IT staff to reboot the VM (tisk tisk) in hopes of solving the problem. The CPU usage was apparently pegged at 100% and the system became unresponsive.
This can't be a disk issue due the the horsepower of our SAN. This leaves CPU and memory, possibly Apache and mysql tuning, or incompatibilities between/amongst the various verions of our LAMP components. The VM had 512MB at the time and as I said a single CPU.
Just today I increased the CPU count to 4, replaced the kernel with the SMP kernel (do NOT use the BIGSMP kernel), and bumped the memory to 1GB. I locked this VM to ESX blade 1 to keep DRS from vmotioning it. I increased the number of pre-fork httpd processes in the Apache2 config from 10 to 20 and max idle to 40. I should know this coming week if it helped. I'm researching the chat performance here on the forum. My gut instinct is that I'm probably far better off running a dedicated chat server on another VM. I'm no Apache or mysql guru so there's probably still much performance for me to gain by tweaking these daemons.
Oh, and btw, ESX simply rocks, especially in a blade server and SAN environment. We run most of our back end servers on the two blades mentioned above, with the exception of our Exchange server, MS sql server, and 4 Citrix servers. We have both our W2K3 AD DCs and about 5 other windows servers running as VMs as well as all of our Linux servers--Novell iFolder and ZLM servers, Moodle, as well as a Debian Request Tracker (helpdesk) server. All of our VMFS volumes, VMDKs, and their mapped LUNs (RDMs) reside on the SAN arrays, one IBM FasTt 600 and one Nexsan Satablade. The performance and reliability we've experienced with ESX is phenominal.
Stan Hoeppner
Sysadmin
Whitfield School
Interesting question. It's one I've just spent quite a bit of time recently trying to answer for myself.
My environment:
IBM Bladecenter model 8677
IBM Bladeserver LS20 dual CPU dual core 2 GHz Opteron (4 cores)
8GB RAM
200 MB/s fiber channel SAN
Nexsan array, 512MB cache, 8x500GB 7200RPM SATA drives in a RAID5 pack
VMware ESX 3.0 Enterprise
We have two of these blades and run about 14 VMs across the two, with a massive amount of CPU cycles to spare. We rarely run more than 10-15% CPU utilization across both the blades. Memory usage runs about 4-5GB per blade.
We've been running an ESX VM of SuSE Linux Enterprise Server 10, Moodle 1.7, mysql 5.x, Apache2, and php5. I had been using the uniprocessor 2.6.27.x kernel and only 1 CPU core. One of our faculty had his class using the chat server on Friday. Unfortunately I wasn't involved at the time, but I'm told the system hit the wall prompting a less experienced member of our IT staff to reboot the VM (tisk tisk) in hopes of solving the problem. The CPU usage was apparently pegged at 100% and the system became unresponsive.
This can't be a disk issue due the the horsepower of our SAN. This leaves CPU and memory, possibly Apache and mysql tuning, or incompatibilities between/amongst the various verions of our LAMP components. The VM had 512MB at the time and as I said a single CPU.
Just today I increased the CPU count to 4, replaced the kernel with the SMP kernel (do NOT use the BIGSMP kernel), and bumped the memory to 1GB. I locked this VM to ESX blade 1 to keep DRS from vmotioning it. I increased the number of pre-fork httpd processes in the Apache2 config from 10 to 20 and max idle to 40. I should know this coming week if it helped. I'm researching the chat performance here on the forum. My gut instinct is that I'm probably far better off running a dedicated chat server on another VM. I'm no Apache or mysql guru so there's probably still much performance for me to gain by tweaking these daemons.
Oh, and btw, ESX simply rocks, especially in a blade server and SAN environment. We run most of our back end servers on the two blades mentioned above, with the exception of our Exchange server, MS sql server, and 4 Citrix servers. We have both our W2K3 AD DCs and about 5 other windows servers running as VMs as well as all of our Linux servers--Novell iFolder and ZLM servers, Moodle, as well as a Debian Request Tracker (helpdesk) server. All of our VMFS volumes, VMDKs, and their mapped LUNs (RDMs) reside on the SAN arrays, one IBM FasTt 600 and one Nexsan Satablade. The performance and reliability we've experienced with ESX is phenominal.
Stan Hoeppner
Sysadmin
Whitfield School
Hi Tim,
Don't hesitate to run Moodle under VMware ESX. As long as you have fairly current hardware. After sorting out the chat issue, our moodle server I previously described is running pretty well. Under ESX, be sure to run each VM(guest) as a uniprocessor machine, in other words, no virtual smp. If you run into performance problems, you can always build more VMs to separate the load, i.e. build a dedicated mysql or postgre server, a dedicated apache server for moodle, a dedicated chat server for running chatd.php. You can make the servers painfully small so they don't each much ram in the esx environment, and you'll get fantastic CPU utilization from your hardware. The possibilities for flexibility are endless if you use SAN storage in your ESX environment.
Don't hesitate to run Moodle under VMware ESX. As long as you have fairly current hardware. After sorting out the chat issue, our moodle server I previously described is running pretty well. Under ESX, be sure to run each VM(guest) as a uniprocessor machine, in other words, no virtual smp. If you run into performance problems, you can always build more VMs to separate the load, i.e. build a dedicated mysql or postgre server, a dedicated apache server for moodle, a dedicated chat server for running chatd.php. You can make the servers painfully small so they don't each much ram in the esx environment, and you'll get fantastic CPU utilization from your hardware. The possibilities for flexibility are endless if you use SAN storage in your ESX environment.
Hi Tim,
We are running a Moodle cluster on VMWare ESX running SLES 10. We have extremely beefy hardware (8 core boxes with 16GB Ram) and we get apache slowness problems. In fact at times it has been my experience that our old Xeon single core 2Ghz box has out-performed the ESX cluster.
Im looking for suggestions as well. It seems that there maybe some apache tweaks we put in that may have fixed this intermittent problem.
We have isolated the problem down to the actual GSX/ESX (we have tested both) not responding to GET requests or not forwarding them to the host properly. This can happen under heavy load and under light weight nodes.
If anyone has any suggestions I would love to hear them.
Kindest Regards.
We are running a Moodle cluster on VMWare ESX running SLES 10. We have extremely beefy hardware (8 core boxes with 16GB Ram) and we get apache slowness problems. In fact at times it has been my experience that our old Xeon single core 2Ghz box has out-performed the ESX cluster.
Im looking for suggestions as well. It seems that there maybe some apache tweaks we put in that may have fixed this intermittent problem.
We have isolated the problem down to the actual GSX/ESX (we have tested both) not responding to GET requests or not forwarding them to the host properly. This can happen under heavy load and under light weight nodes.
If anyone has any suggestions I would love to hear them.
Kindest Regards.
we're experiencing a similar situation here.
ESX on a QC 3ghz dual proc machine w/ 32gb.
we were getting sql timeouts w/ only 20-30 users.
cpu <25%, lots of ram available. graphs of network, disk all indicated no stress..
i did notice that each user had between 10-20 connections though.
any default parameters that this may indicate perhaps?
-ams
ESX on a QC 3ghz dual proc machine w/ 32gb.
we were getting sql timeouts w/ only 20-30 users.
cpu <25%, lots of ram available. graphs of network, disk all indicated no stress..
i did notice that each user had between 10-20 connections though.
any default parameters that this may indicate perhaps?
-ams
I wouldn't lean towards the problem with the actual ESX software or the underling hardware. This is more than likely a configuration problem. Perhaps a problem with sessions or with a corrupted table. We run ESX and any problem we've had has looked like software/hardware until we dig deep and determine the problem with the underlining configuration.
A couple examples:
A corrupted quiz table causes mysql load to spike as it reads rows without indexes thus causing apache to backup.
Sessions stored on a NFS server suffer from flocking due to poor NFS implementation. This also causes apache to backup and not respond to various requests.
We've had session problems in mysql database based on adodb library problems, see this thread, http://moodle.org/mod/forum/discuss.php?d=106108
My point here is that I'd look every other place before I'd look at ESX for the problem. Check your mysql slow querry log. Do some strace of your apache threads to see if there are any locks or disk read issues. Other system settings you changed when migrating boxes to ESX? I'd bet there is some underlining cause of the problem that is not hardware or esx releated.
A couple examples:
A corrupted quiz table causes mysql load to spike as it reads rows without indexes thus causing apache to backup.
Sessions stored on a NFS server suffer from flocking due to poor NFS implementation. This also causes apache to backup and not respond to various requests.
We've had session problems in mysql database based on adodb library problems, see this thread, http://moodle.org/mod/forum/discuss.php?d=106108
My point here is that I'd look every other place before I'd look at ESX for the problem. Check your mysql slow querry log. Do some strace of your apache threads to see if there are any locks or disk read issues. Other system settings you changed when migrating boxes to ESX? I'd bet there is some underlining cause of the problem that is not hardware or esx releated.