I did some additional testing today comparing performance of open-iscsi + tgt/rbd vs tgt/ramdisk to verify where the performance bottleneck seems to occur. With a simple RAM disk iscsi target, I could push up to about 1GB/second for both read and write operations. Running the same tests (using fio) against an RBD based iscsi target, I only could get up to about 400MB/sec for both read and write (write was slightly slower). Running yet another test with fio, using the rbd ioengine, I can generate 1GB/sec for reads and about 800MB/sec for writes. tgtd/ramdisk: 1GB/sec 1GB/sec (read/write) tgtd/rbd: 400MB/sec 388 MB/sec librados: 1GB/sec 800MB/sec All of these tests were against a tgtd with 512 threads. The fio job parameters are as follows, I only varied the ioengine, rw, and filename settings depending on what was being tested. [default] rw=randread size=10g bs=1m ioengine=libaio direct=1 numjobs=1 filename=/dev/sdb runtime=600 write_bw_log=iscsiread iodepth=256 iodepth_batch=256 So, open-iscsi is certainly capable of better speed, and librados can get better speed, so the bottleneck appears to be either in tgtd or in the context switch between the iscsi initiator and tgtd. I wanted to share this with the devs here, I'll keep looking for other areas of improvement. thanks, Wyllys -- 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
