Running 300 diskless clients off of one server.
Server <=4G bond0=> core-switch <=1G => lab-switch <=100M=>clients.
I'm looking to optimize the network traffic a bit more by using squashfs + nbd. I was wondering if there are some recommendations for block size optimizations or if I'm just going to have to do my own bench marks?
nbd-client: -block-size <block_size> Use a blocksize of "block size". Default is 1024; allowed values are either 512, 1024, 2048 or 4096
mksquashfs: -b <block_size> set data block to <block_size>. Default 131072 bytes
With the current NFS file system the majority of files are less than 64K so 64K rsize & wsize was optimal ( trial tests ). 32K and 64K+ degrade performance.
Here is what I think :
With nbd-client you should care about the MTU of your network. 1024 seems good.
Squashfs is using cache, so perfomance should be good with 131072 octets (as compression will be better). Of course, this depends on the number of hits with the cache. If your clients have lot of RAM, the cache will be bigger and the number of cache hits should be high.