Yes, you're right. The speed of an individual write would be faster with the global lock and what I should have said is that we could improve our overall throughput if we allow concurrent threads to write at the same time.
I will try running the test suite against it. BTW, do you know whether the test suite has been changed or added to at all between 2.1 M1 and the 2.1 beta? Thanks, Warwick ----------------------------------------------------------- Warwick Burrows E2open Senior Engineer 9600 Great Hills Trail, #325 http://www.e2open.com Austin TX 78759 ----------------------------------------------------------- -----Original Message----- From: Oliver Zeigermann [mailto:[EMAIL PROTECTED] Sent: Tuesday, August 03, 2004 10:14 AM To: Slide Users Mailing List Subject: Re: Concurrent upload error: Status Quo Warwick Burrows wrote: >>>The default 2.1 distribution will not have any deadlock problems. > > However, at the cost of a single write request at a time. > > I guess serializing writes is a result of the global lock you talk > about implementing in the original "Concurrent upload error: Status > Quo" thread. Would you mind explaining the problem to us so that we > can also understand the reasoning behind serializing writes. I'm > guessing its because the history directory is the choke point. Exactly. I explained this in another thread, but have somewhat lost oversight... > I ask because our Slide client uses its own authorization and doesn't > use Slide security so we have turned it off. It also doesn't need > auto-versioning because the client will handle file versioning > explicitly. So if I enable all four of the workarounds you had > mentioned to avoid deadlocks then can I turn off the global write lock > in the 2.1 beta and increase my write-speed? Why not give it a try? Try running with testsuite with 50 threads or so. If you do not experience *any* deadlocks it looks good. Might be possible with TLocks and Sequences. I do not know. However, in no case will the write speed be improved. Tests with the global read/write lock are the fastest. What is increased is the possible concurrency. Oliver --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
