HP Tru64 UNIX and TruCluster Server Version 5.1.B-4 Patch Summary and Release Notes (13156)
You can minimize the likelihood of these problems as follows:
• Avoid congestion on your LAN and cluster interconnect.
• Ensure your servers have enough excess capacity to respond quickly to NFS requests that
modify the file system (writes, file and directory creation, and so forth.)
• Increase the size of the server's duplicate request cache when the nfsstat command shows
a large number of retransmits to clients. For instructions on increasing the size of the cache,
see “Tuning the NFS Server Duplicate Request Cache”.
You can monitor the number of NFS retransmissions using the nfsstat -c command. The
retrans field indicates the number of retransmissions. A retransmission rate higher than 2%
indicates a potential problem.
The following example shows the output from the nfstat -c command. The retransmission
fields are marked with asterisks (*). This example is of a client workstation in a typical
environment.
% nfsstat -c
Client rpc:
tcp: calls badxids badverfs timeouts newcreds
0 0 0 0 0
creates connects badconns inputs avails interrupts
0 0 0 0 0 0
udp: calls badxids badverfs timeouts newcreds *retrans*
224518870 959 0 101985 0 0
badcalls timers waits
102013 110894 0
Client nfs:
calls * retrans* badcalls nclget nclsleep ndestroys ncleans
224414222 4248 28 224414282 0 6219 224408063 \
If an overwhelmed duplicate request cache condition occurs, we recommend you perform one
or more of the following tasks:
• Ensure that there are short periods of idle time on the I/O subsystem and network links.
• After a file is written, do not rewrite it for a few minutes.
• Delete and recreate files instead of overwriting the same file repeatedly.
• Use Memory Channel cluster interconnect.
To avoid overwhelming the duplicate request cache:
• Do not run hundreds of simultaneous processes that write files
• Do not operate the system under so heavy a load that NFS operations frequently take several
seconds to complete.
Use the netstat command to determine if your network is saturated. For Ethernet networks,
a high number of collisions indicates that the network may be saturated. The following example
shows the output from the netstat -I tu0 command:
Name Mtu Network Address Ipkts Ierrs Opkts Oerrs *Coll*
tu0 1500 <Link xx:xx:xx:xx:xx 840386045 0 254319298 5121 5014223
tu0 1500 network client 840386045 0 254319298 5121 5014223
tu0 1500 DLI none 840386045 0 254319298 5121 5014223
3.2.2.13 Tuning the NFS Server Duplicate Request Cache
The NFS server maintains a list of recently completed nonrepeatable requests. This list is used
to reply to client retransmissions of the request in the event that the initial request transmission's
reply was lost or that the server took too long to satisfy the request.
3.2 Prior Release Notes 53