Dale Ghent wrote: > On Nov 15, 2005, at 6:27 PM, Jim Rees wrote: > >> CheckHost in fact already runs with the host held. I was wondering >> whether >> it was really necessary to keep locking and unlocking the hash table but >> decided to play it safe for now and leave it as-is. > > Doing some research into lock debugging on Solaris, I came across a new > utility on Solaris 10 which uses Dtrace to instrument a running process. > > man plockstat > > I'll try using it if (when) things hang here again. > > Here's an example of it running against the fileserver process on one of > our AFS servers: > > [EMAIL PROTECTED]>plockstat -p 15036 > ^C > Mutex block > > Count nsec Lock Caller
Do the counts and nsec columns indicate total number of calls and time wait spent since the beginning of the process? Since plockstat was called? Or something else? Assuming it is total number of calls and time waited it calls into question some of the theories of the impact of the host mutex lock contention being the problem since the wait values are very small compared to the count. Please provide more detail on what these numbers mean. Jeffrey Altman
smime.p7s
Description: S/MIME Cryptographic Signature
