| |||||||||||||||||||||||||||||||
|
Christopher εφη Also please note that the network interfaces that use 9000 MTU packets are incompatible with everything else on your lan , so you _SHOULD_ use a secondary or tertiary adapter , whatever the case , for connecting your NFS network together! Now I was thinking about my previous post about storage performance for an NFS server so I found out an nfstat utility , and on one of my maildir servers it says Server nfs v3: null getattr setattr lookup access readlink 353 0% 919031723 33% 163258318 6% 333941651 12% 655894862 24% 36 0% read write create mkdir symlink mknod 251144060 9% 192318738 7% 14792049 0% 523767 0% 414 0% 1 0% remove rmdir rename link readdir readdirplus 17355778 0% 5628 0% 11622038 0% 6778998 0% 8314094 0% 133659263 4% fsstat fsinfo pathconf commit 330150 0% 374 0% 0 0% 3717009 0% so setattr and write are only 13% of the total NFS work which denotes that what is needed for a good email NFS server is oodles or ram for the attribute caches, and fast reads i.e. striped underlying storage. >> >> >> When running NFS over a ~ 9000 byte MTU network I normally stick to an >> 8192 r/w size. Some NFS stacks have issues with high volume user where >> requests span multiple packets. >>
| ||||||||||||||||||||||||||||||
© 2004-2008 readlist.com