[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Understanding nfsd thread stats



The NFS Howto is pretty cryptic about how to read nfsd thread
statistics. AFAICT, if I have the following output:

    $ cat /proc/net/rpc/nfsd
    rc 0 832888 1838270
    fh 0 0 0 0 0
    io 1230212044 2485604185
    th 16 205835 4865.996 293.876 452.380 4.160 166.324 138.216 118.876 126.344 6.620 857.204
    ra 32 1748589 26409 6556 81 5 22 20 4 15 7 2177
    net 2671200 5 2671087 406
    rpc 2671158 0 0 0 0
    proc2 18 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    proc3 22 13 20235 478 21496 7504 9 1783885 830080 836 19 0 0 974 6 495 0 502 120 74 6 0 3096
    proc4 2 0 0

The last ten entries on the th line are supposed to be "deciles" which
presumably tell me how much work is being done at each percent of thread
usage in 10% increments. Bizarre, and poorly-explained, but okay. What's
not obvious to me is whether the "top three deciles" are the first three
on the left, or the last three on the right.

In other words, have my threads been maxed out for 4865.996 seconds, or
857.204? And is there a better idea to determine if I need more (or
fewer) nfsd threads?

-- 
Unabashedly littering the information superhighway with detritus like
this for over 15 years now.



Reply to: