| Setup: |
| |
| Target: 2.6.29 kernel, 64 bit, Intel E8400 CPU @ 3.00GHz, 4 GB RAM, SCST trunk |
| revision 727 (which is close to the 1.0.1 release). A file of 1 GB residing |
| on a tmpfs filesystem has been exported via SCST. |
| |
| Initiator: 2.6.29 kernel, 64 bit, Intel E6750 CPU @ 2.66 GHz, 2 GB RAM, |
| openSUSE 11.0 userspace. |
| |
| Network: two MHGH28-XTC (MT26418) ConnectX InfiniBand HCA's connected back to |
| back, which are DDR PCIe 1.0 HCA's. The IPoIB stack was configured with the |
| default MTU of 2044 bytes on both interfaces and was using datagram mode. |
| ib_read_bw reported a throughput of 1394 MB/s for this network, and netperf |
| reported a TCP/IP throughput of 1200 MB/s (with default parameters). |
| |
| Results: |
| |
| Buffered I/O, block size of 512K (dd if=/dev/sdb of=/dev/null bs=512K): |
| |
| write-test: iSCSI-SCST 243 MB/s; IET 192 MB/s. |
| read-test: iSCSI-SCST 291 MB/s; IET 223 MB/s. |
| |
| Buffered I/O, block size of 4 KB (dd if=/dev/sdb of=/dev/null bs=4K): |
| |
| write-test: iSCSI-SCST 43 MB/s; IET 42 MB/s. |
| read-test: iSCSI-SCST 288 MB/s; IET 221 MB/s. |
| |
| Or: depending on the test scenario, SCST transfers data between 2% and |
| 30% faster via the iSCSI protocol over this network than IET. |
| |
| Note: at least for the tests with a block size of 4 KB, the initiator |
| system was the bottleneck, not the target system. |
| |
| Something that is not relevant for this comparison, but interesting to |
| know: with the SRP implementation in SCST the maximal read throughput |
| is 1290 MB/s on the same setup. |
| |
| Measured by Bart Van Assche |
| |