I queried it from the ASM instance. Delta read time / Delta reads over a few minutes. I didn't correlate against other metrics but the times seem reliable. Harel Safra Sent from my phone On Dec 8, 2010 7:46 PM, "Cary Millsap" <cary.millsap@xxxxxxxxxxxx> wrote: > Is the AVG_WRITE_TIME statistic actually reliable? I think it's probably > not—at least it wasn't some time in the past, if you were using async I/O. > > Cary Millsap > Method R Corporation > http://method-r.com > > > On Wed, Dec 8, 2010 at 11:41 AM, Harel Safra <harel.safra@xxxxxxxxx> wrote: > >> On 08/12/2010 17:13, Amaral, Rui wrote: >> >> Yes in theory that can be true but it depends on several factors: >> >> >> Why in theory? Please have a look at a snapshot of statistics from one >> node of our data warehouse cluster running on EMC DMX4 storage: >> *AVG_WRITE_TIME* *TOTAL_WRITES* *AVG_READ_TIME* *TOTAL_READS* 1.1 40460 >> 26.78 35864 >> Much better write times than read times (the high read time is also >> contributed from the large IO the database tends to do). >> >> 1) assuming write-back caches on the array have not been disabled (which in my opinion they should be) >> >> Why do you think write back cache should be disabled? Proper storage >> systems have sufficient battery backups built in to flush the whole cache to >> disk in case of power failure. >> >> Harel Safra >>