From mboxrd@z Thu Jan 1 00:00:00 1970 From: Joel Becker Date: Wed, 1 Dec 2010 21:30:55 -0800 Subject: [Ocfs2-devel] [PATCH 2/2] Thread recovery In-Reply-To: <4CF72B01.4000500@oracle.com> References: <4CF72B01.4000500@oracle.com> Message-ID: <20101202053055.GH16604@mail.oracle.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ocfs2-devel@oss.oracle.com On Wed, Dec 01, 2010 at 09:13:37PM -0800, Sunil Mushran wrote: > One problem I see is that we may spawn too many threads. Imagine > a setup with 50 mounts and 16 nodes. Not uncommon at all. If 5 nodes > die, that's 300 threads, 50 of which will be coordinating threads. > > One solution is to have a universal mount count and use that and the > number of cpus to come up with a per-mount parallel reco count. > > Thoughts, anyone? The new concurrency workqueues should allow us a limited-threadcount pool of threads to do this. We should be using them rather than implementing thread management ourselves. Joel -- "Three o'clock is always too late or too early for anything you want to do." - Jean-Paul Sartre Joel Becker Senior Development Manager Oracle E-mail: joel.becker at oracle.com Phone: (650) 506-8127