[lustre-discuss] question about /proc/fs/lustre/mds/MDS/mdt/stats

[Dilger, Andreas]

"samples" is the number of times this event was measured, and it happens that these request stats are all measured together.  The values to the right of the units are min/max/sum/[sumsq] in units of microseconds, or seconds, or requests.  To work out the average request waittime is:

       (sum / samples) = 1257652419050 / 611801704 = 2055usec

Cheers, Andreas
-- 
Andreas Dilger

Lustre Principal Architect
Intel High Performance Data Division

On 2016/09/16, 10:45, "Mahmoud Hanafi" <mahmoud.hanafi at nasa.gov> wrote:

    Why is 'samples' the same for req_waittime, req_qepth, req_active, 
    req_timeout all the same.
    
    req_waittime              611801704 samples [usec] 3 5061774 
    1257652419050 953071357361477866
    req_qdepth                611801704 samples [reqs] 0 89 21110263 22332201
    req_active                611801704 samples [reqs] 1 127 2647563677 
    122463789917
    req_timeout               611801704 samples [sec] 150 150 91770255600 
    13765538340000
    reqbuf_avail              1256133286 samples [bufs] 481 574 705260598158 
    396165104527476
    ....
    
    shouldn't the samples for [usec] be total usec similar to units of 
    [bytes]. For example:
    cat /proc/fs/lustre/obdfilter/nbp8-OST00d1/stats
    snapshot_time             1474043858.153583 secs.usecs
    read_bytes                21704638 samples [bytes] 4096 1048576 
    12917847764992
    write_bytes               25971348 samples [bytes] 1 1048576 19719890405713
    ....
    
    Samples here is total bytes transferred correct?
    
    -- 
    Mahmoud Hanafi
    
    _______________________________________________
    lustre-discuss mailing list
    lustre-discuss at lists.lustre.org
    http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org