Performance problem on NFS server
2007-12-24 18:54:00
The system under question is an Ultra 5500 with 4 cpu running
Solaris 2.6 patched up. It is an NFS Server which accommodate a
few hundred users' home and some other exported directories.
Users are complaining of system being sluggish and slow in
respond. The following is the output of "se zoom.se" for the red
area only. I have already implemented DNS rr on the NIC which
helped that problem a little. MUTEX is what puzzles me.
As always, I will summarize.
NFS CLIENT INFORMATION
The current state of this NFS client is:
red Packets being lost, fix bad network or reduce NFS/RPC
packet sizes
NFS client RPC call rates and totals:
Calls : 429496729.60/s 33686
Timeouts : 429496729.60/s 20586
Badxids : 0.00/s 0
CLIENT RPC RULE THRESHOLDS
RPCCLIENT_MINCALLS default= 0.1/s getenv= 0.1/s calls per sec
for idle
RPCCLIENT_TIMEOUT default= 5.0% getenv= 5.0% calls that
timeout
RPCCLIENT_BADXID default= 0.0% getenv= 0.0% timeouts with
badxid
----------------------------------------------------------------
--------
GGREGATE NETWORK INFORMATION
Overall state of the network interfaces:
red Add more or faster nets
IP dropping incoming packets
State Name Ipkt/s Opkt/s Err/s Coll% NoCanPut/s Defer/s
red hme0 7456.5 7104.7 0.00 0.00 1.51 0.00
blue hme1 0.0 0.0 0.00 0.00 0.00 0.00
blue hme2 0.0 0.0 0.00 0.00 0.00 0.00
blue hme3 0.0 0.0 0.00 0.00 0.00 0.00
TCP measures Total Rate/s
Input bytes 10059218 2011843.60
Output bytes 4788797 957759.40
Retransmit bytes 0 0.00%
Connections 4 0.80
ETHERNET RULE THRESHOLDS
ENET_INUSE default=50.0/s getenv=50.0/s ignore fewer
output packets/s
ENET_COLL_WARNING default=15.0% getenv=15.0%
ENET_COLL_PROBLEM default=30.0% getenv=30.0%
ENET_ERROR_PROBLEM default=0.050/s getenv=0.050/s
ENET_DEFER_WARNING default=2.00% getenv=2.00% some delayed
output packets
ENET_DEFER_WARNING default=10.00% getenv=10.00% too many
delayed output packets
ENET_NOCANPUT_PROBLEM default=1.000/s getenv=1.000/s incoming
packet dropped by IP
----------------------------------------------------------------
---------
MUTEX INFORMATION
The current mutex state is:
red Serious mutex contention, multiprocessor kernel
overloaded,
use the lockstat(1M) command to diagnose the cause,
faster CPUs will help, more CPUs will make it worse
Total number of spins on mutexes: 60657 per second
Average spins on mutexes per CPU: 15164 per second
MUTEX RULE THRESHOLD
MUTEX_BUSY default=500.0/s getenv=500.0/s OK up to
this level per-CPU
MUTEX_SYS_PROBLEM default=20.0% getenv=20.0% OK up to this
amount of system CPU time
MUTEX_USR_SYS_PROBLEM default=1.0ratio getenv=1.0ratio OK if
usr/sys ratio is higher than this
________________________________________________
Get your own "800" number
Voicemail, fax, email, and a lot more
http://www.ureach.com/reg/tag
Comments
Got something to say?
You must be logged in to post a comment.
