On 11/08/2011 03:46 AM, FUJITA Tomonori wrote: > > Batching commands sounds good. I've not read the patches though. Have > you measured that how this could improve the performance? Yes, i have measured the performance. Improvements depend on the workload type, the storage device characteristics, and the type of the servers. The main improvement is for IOPS. I saw (over iser/ib to FusionIo SSD) improvement of 5-20% for write and 3-15% for read IOPS (1kb) for various loads/setups. Typical values 55Kiops -> 62Kiops, 90kiops -> 100kiops iser can enjoy batching because it repeatedly polls the Completion Queue and may found multiple cmds in one polling sweep. It stores them on a list and instantly processes it and submits the commands. Here it treats the commands which already present on the list as a batch, so the processing is never actually deferred in order to get some more. Only if some commands are "already here" anyway, are they marked as a batch. Do you see how other transports, notably iscsi/tcp can employ a similar scheme? -- 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
