Comment 54 for bug 561210

Revision history for this message
Alex (d-f0rce) wrote :

@Andy Whitcroft

Your sysctl setting on its own did not solve the problem for me. However while I was googling for vm.min_free_kbytes to check out what it actually does, I came across this site: http://russ.garrett.co.uk/2009/01/01/linux-kernel-tuning/

So I set these values on both, the server and the client:

## increase amount of kernel memory
vm.min_free_kbytes = 65536

## increase TCP max buffer size
net.core.rmem_max = 16777216
net.core.wmem_max = 16777216

## increase TCP autotuning buffer limits
net.ipv4.tcp_rmem = 4096 87380 16777216
net.ipv4.tcp_wmem = 4096 65536 16777216

With these settings my problem is gone using the current lucid kernel 2.6.32-24.42. The system still gets a bit laggy on very high NFS network load but it does not stall anymore.