nevertheless-it-spins
July 10th, 2011, 10:02 AM
Hi everyone
Ubuntu 10.10
We have an Ubuntu cluster made on the basis of fat clients of ltsp. The cluster has been working for several months on a series of 4-core nodes.
Now we installed a new system on the basis of 4-socket/12-core Supermicro platform (http://www.supermicro.nl/Aplus/system/2U/2042/AS-2042G-TRF.cfm) and a problem appeared. See attached screenshot.
But I do not think that this is a hardware problem. All hardtests are OK.
It seems to me the problem related to the storage structure I chose. It is the following.
Rootfs is aufs, where the ro-filesystem is nfs and rw-filesystem is either tmpfs or nfs4 on a local disk. Moreover, nfs is mounted using cachefiles. I know it is a bit complcated, but it worked on all other systems.
I guess the problem appears due to massive simultaneous read-write whick appears with about 12 tasks.
Could you offer some solutions?
Ubuntu 10.10
We have an Ubuntu cluster made on the basis of fat clients of ltsp. The cluster has been working for several months on a series of 4-core nodes.
Now we installed a new system on the basis of 4-socket/12-core Supermicro platform (http://www.supermicro.nl/Aplus/system/2U/2042/AS-2042G-TRF.cfm) and a problem appeared. See attached screenshot.
But I do not think that this is a hardware problem. All hardtests are OK.
It seems to me the problem related to the storage structure I chose. It is the following.
Rootfs is aufs, where the ro-filesystem is nfs and rw-filesystem is either tmpfs or nfs4 on a local disk. Moreover, nfs is mounted using cachefiles. I know it is a bit complcated, but it worked on all other systems.
I guess the problem appears due to massive simultaneous read-write whick appears with about 12 tasks.
Could you offer some solutions?