Andrew, Actually no I havn't had much experience with larger journals. I've basically only been testing for the past few days, and hadn't gotten that far yet. Would you say that for this type of application a journal size of about 200MB would be appropriate for partitions that are heavily written to? Would this need to be greater than 200 in any case? Currently we're only using dual 18GB SCSI drives in most machines. One of my other main concerns is quotas. I hear ext3+quota = deadlocks in alot of cases. Luckily I havn't seen any as of yet. I'm hoping that I don't see any at all. I'm assuming that using the data=journal mode is beneficial to quotas on a server as well. Thanks again for your input. -Mark On Thursday 01 May 2003 02:12 am, Andrew Morton wrote: > Mark A Basil <mbasil@alabanza.com> wrote: > > Thanks, Andrew. > > > > Most webservers are, in fact, mostly read-only. But these are more of a > > complete hosting solution, so they are web/mail/pop/sql servers. I've > > run about 8 different benchmarking tests, 4 per ordered/journal modes, > > and journal wins out by far. The times for reads/writes in journal'd > > mode were nearly 1/4th of those in ordered. > > Ah, OK. > > I assume you have experimented with large journals? That seems to always > be a win in that sort of situation. > > > Also, I was testing with the elevator settings at > > read = 16384 and write = 8192 > > > > So, I'm going to push for data=journal if I am positive that the sync > > bug had been fixed. Any ideas on that? > > It's fixed. > > > > > _______________________________________________ > > Ext3-users@redhat.com > https://listman.redhat.com/mailman/listinfo/ext3-users _______________________________________________ Ext3-users@redhat.com https://listman.redhat.com/mailman/listinfo/ext3-users
