> -----Original Message----- > From: linux-rdma-owner@xxxxxxxxxxxxxxx [mailto:linux-rdma- > owner@xxxxxxxxxxxxxxx] On Behalf Of Ming Lin > Sent: Sunday, May 01, 2016 1:07 AM > To: Sagi Grimberg; linux-rdma@xxxxxxxxxxxxxxx > Cc: Bart Van Assche; target-devel > Subject: Re: iSER multi-queue configuration > > On Sun, 2016-05-01 at 05:53 +0300, Sagi Grimberg wrote: > > > Hi Sagi, > > > > Hi Ming, > > > > > I'm looking at why there is only 1 HW queue for iser. > > > > > > 2185 int scsi_mq_setup_tags(struct Scsi_Host *shost) > > > 2186 { > > > ... > > > 2199 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : > > > 1; > > > > > > > > > It seems because iser code does not set shost->nr_hw_queues at all. > > > > > > Why? > > > > The iscsi driver has not yet been converted to scsi_mq and iser as > > an iscsi transport still does not use multi-queue. > > I'm going to compare performance between NVMe over RDMA and SCSI over > RDMA. SRP supports scsi_mq, so I'd use SRP to do the apples-to-apples > comparison. > > > > > In iscsi, a hw queue maps to an iscsi session. In order to support > > multi-queue we need to have iscsi open multiple sessions and map > > a session to a queue. > > > > I had some preliminary works on this (joint with Mike Christie) but I > > didn't have time thus far to implement a full conversion to scsi-mq. > > I'd happy to test it once you done. Ditto. SRP as a comparison is only useful for IB/RoCE transports. iSER will allow iWARP to play in the game as well... Steve. -- To unsubscribe from this list: send the line "unsubscribe target-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
