> Well we are actually getting better writes from tcp, which is
weird
> due to the small overhead that tcp would have. Here is what I
have
> currently set on a 2.8 client:
if you obtain better perf on tcp, i guess u perhaps
lost packets on udp and u should then check ur network : check route from one
machine to another with traceroute for example and try to examine where it could
possibly cogestion or missroute beetween filer and solaris station (if u have a
really wide network, it is normal to loose packets on udp so u were right to be
on tcp)
> The nfs ops have increased, the
only thing I do see if I do a large
> cvs checkout the writes are
intermitent. I see zeros and then I see a
> large write, below is an
example. Now is this due to the filer
> flushing what it has on memory to
disk?
yes this is normal, the filer cache writes in the
nvram then writes in a singlke shoot on disks