High CPU time from nfsd threads in svc_poll() (Doc ID 2114369.1)

Last updated on JANUARY 24, 2017

Applies to:

Solaris Operating System - Version 8 6/00 U1 to 10 1/13 U11 [Release 8.0 to 10.0]
Information in this document applies to any platform.

Symptoms

 An NFS server's nfsd daemon is consuming very large amounts of CPU - at times when the NFS client workload is not abnormally high. 

An mpstat indicates very high contention (smtx), and lockstat reveals most of the CPU time is in svc_poll(), svc_xprt_qget() or svc_squeuereq() calls - for example:

Adaptive mutex spin: 234513 events in 5.140 seconds (45626 events/sec)
Count indv cuml  rcnt nsec    Lock           Caller
----- ---- ----- ---- ------- -------------  ------------------------------------
71114  30%  30%  0.00 4162228 0x3036584cae0  cv_timedwait_sig_hires+0x1c0
58276  25%  55%  0.00 1159092 0x3036584cae0  svc_poll+0x1b8
55601  24%  79%  0.00 1033661 0x3036584cae0  svc_poll+0xd8
31141  13%  92%  0.00    9251 0x6005073ef00  clock_tick_process+0xc8
 5805   2%  95%  0.00  189237 0x3036584cb48  svc_xprt_qget+0xc
 2486   1%  96%  0.00 1613775 0x3036584cae0  svc_queuereq+0x20
 2032   1%  97%  0.00    6110 0x30007409940  clock_tick_process+0x2c
 1144   0%  97%  0.00  502410 0x3036584cae0  svc_poll+0x158
  995   0%  97%  0.00  536363 vx_worklist_lk vx_worklist_process+0x98
  390   0%  98%  0.00  469994 0x3036584cae0  svc_poll+0x9c
... 
  

Cause

Sign In with your My Oracle Support account

Don't have a My Oracle Support account? Click to get started

My Oracle Support provides customers with access to over a
Million Knowledge Articles and hundreds of Community platforms