Sorry, I missed the list in my previous reply... On Thu, Aug 28, 2008 at 2:09 AM, FUJITA Tomonori <fujita.tomonori@xxxxxxxxxxxxx> wrote: > On Wed, 27 Aug 2008 20:22:42 +0300 > "Alexander Nezhinsky" <nezhinsky@xxxxxxxxx> wrote: >> On Wed, Aug 27, 2008 at 8:01 PM, FUJITA Tomonori >> <fujita.tomonori@xxxxxxxxxxxxx> wrote: >> >> > Thanks. But you can do faster. This patch incurs the thread and pipe >> > notification overheads. You don't need bs_thread stuff. See bs_aio.c >> >> Sure. I intended to do it that way later. I'm going to send some performance >> results over ISER in a couple of days, so i first used this >> straightforward implementation. > > Nice. > >> Moreover, it is also interesting to measure *with* the thread'n'pipe >> overhead because >> this way you only exclude I/O. It can be later compared with a version >> without threads >> to assess their impact. > > Yeah, the results should be interesting because we plan to remove the > thread mechanism (well, we need to wait for the AIO support in Linux > kernel). > >> > After that, you can do more faster. You can use sendfile (send data >> > without memory copy) though you need some changes to the core code. >> >> How is sendfile() applicable to this null i/o backing store? I am not >> sending anything >> to I/O anyway, so there is no memory copy to spare. > > You send data to an initiator without doing disk I/Os. > > Anyway, you are an iSER guy, so you can forget sendfile stuff. You mean sendfile /dev/zero --> socket or socket --> /dev/null in case of iSCSI/TCP, don't you? -- To unsubscribe from this list: send the line "unsubscribe stgt" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
