Performance of small sequential non-contiguous writes over NFS suffers when the client has the file locked

Issue

• Customer has a homegrown benchmark that basically enqueues 50,000 messages into a MQseries queue manager queue, and then reads them back out. The load phase is taking 2 to 3 times as long on a RHEL client as it previously did on an AIX client.
• MQseries performance benchmark taking twice as long to write to an NFSv4 export from RHEL than it does from AIX.
• Performance of small sequential non-contiguous writes over NFS suffers when the client has the file locked