On Oct 3, 2007, at 10:31 AM, Roch - PAE wrote: > If the DB cache is made large enough to consume most of memory, > the ZFS copy will quickly be evicted to stage other I/Os on > their way to the DB cache. > > What problem does that pose ?
Personally, I'm still not completely sold on the performance (performance as in ability, not speed) of ARC eviction. Often times, especially during a resilver, a server with ~2GB of RAM free under normal circumstances will dive down to the minfree floor, causing processes to be swapped out. We've had to take to manually constraining ARC max size so this situation is avoided. This is on s10u2/3. I haven't tried anything heavy duty with Nevada simply because I don't put Nevada in production situations. Anyhow, in the case of DBs, ARC indeed becomes a vestigial organ. I'm surprised that this is being met with skepticism considering that Oracle highly recommends direct IO be used, and, IIRC, Oracle performance was the main motivation to adding DIO to UFS back in Solaris 2.6. This isn't a problem with ZFS or any specific fs per se, it's the buffer caching they all employ. So I'm a big fan of seeing 6429855 come to fruition. /dale _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
