RHEL5.7: NFSv4 client hang while database write load simulator running
Issue
- Test to simulate the behaviour of DB2 High Performance Unload, used for backup of the database.
- High load on the NFS share cause the system to hang on regular basis, mostly during the night shortly after the DB unload process is started.
- Brief description of the test case created to simulate real world load: Write to a single 2GB file with 30-120 threads simultaneously in 10MB chunks with o_sync
- Test case runs fine with NFSv3 but hangs with NFSv4
- Test case opens the same file many times and writes to different areas of the file.
- Bad NFSv4 performance with workload combining lock operations with IO, and a tcpdump shows a lot of
NFS4ERR_OLD_STATEIDerrors.
Environment
- Red Hat Enterprise Linux 5.7
- kernel-2.6.18-274.17.1.el5
- System has 48 GB of RAM and 8 cores
- NFSv4 client EMC Datamover NFSv4 server
- Multi-threaded application which repeatedly does WRITEs to a single file while opening it
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.
