On Fri, Jul 23, 2010 at 11:13 AM, Jeff Moyer <jmoyer@xxxxxxxxxx> wrote: > "Ted Ts'o" <tytso@xxxxxxx> writes: > >> On Wed, Jul 07, 2010 at 09:53:30AM +0200, Lukas Czerner wrote: >>> >>> Hi all, >>> >>> since my last post I have done some more testing with various SSD's and the >>> trend is clear. Trim performance is getting better and the performance loss >>> without trim is getting lower. So I have decided to abandon the initial idea >>> to track free blocks within some internal data structure - it takes time and >>> memory. >> >> Do you have some numbers about how bad trim actually might be on >> various devices? > > I'll let Lukas answer that when he gets back to the office next week. > The performance of the trim command itself varies by vendor, of course. > >> I can imagine some devices where it might be better (for wear >> levelling and better write endurance if nothing else) where it's >> better to do the trim right away instead of batching things. > > I don't think so. In all of the configurations tested, I'm pretty sure > we saw a performance hit from doing the TRIMs right away. The queue > flush really hurts. Of course, I have no idea what you had in mind for > the amount of time in between batched discards. It was my understanding that way back when, Intel was pushing for the TRIMs to be right away. That may be why they never fully implemented the TRIM command to accept more than one sectors worth of vectorized data. (That is still multiple ranges per discard, just not hundreds/thousands.) Along those lines, does this patch create multi-sector discard trim commands when there is a large list of discard ranges (ie. thousands of ranges to discard)? And if so, does it have a blacklist for SSDs like the Intel that don't implement the multi-sector payload part of the spec? Greg -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
