We have high i/o wait queue (with idle cpu) on a ultra enterprise 5000
with 512mb ram and many gb of disk space in a storage array; currently,
'ncsize' is set to 30.000 and I see from 'sar' that the directory lookup
cache is not big enough, there're a lots of cache miss; this is an nfs
server with big filesystems (e.g. 10gb) and lots of files in them (e.g.
3 to 5 million); what I'd like to know is:
a. how much memory are we going to pay for incresing 'ncsize' ?
b. is 'bufhwm' of any use (currently unsetted, if I look in the
running kernel is 0)
c. what about *decreasing* 'ncsize' (see bug report 1126134, but
prob. it applied to sunos 4.1.3 only)
Please post a reply to my e-mail address too; thanks,
10. kernel params