Whitepaper: How does iSCSI perform relative to Fiber Channel?
by The editorial team
Contributed by D. John Shakshober
Linux customers often ask how iSCSI performs relative to Fiber Channel. The attached study analyzes the throughput of the Red Hat® Enterprise Linux® 4 iSCSI software initiator w/ 2 1-Gbit nics using an EqualLogic’s PS 3800XV and compares it to a single 2Gbit Fiber Channel Qlogic card running low-level I/O workloads and also running an industry standard Oracle 10g OLTP workload.
Red Hat Linux v4 I/O Performance case study (PDF)
Linux Performance with software iSCSI initiator and EqualLogic storage arrays approaches 2 Gbit Fiber-channel arrays running an Oracle 10g OLTP industry standard workload.
We also share some best practices in configuring multiple 1 Gbit cards for use with EqualLogic iSCSI storage arrays with Red Hat Enterprise Linux 4 U4.
March 29th, 2007 at 11:32 pm
Thanks, very interesting paper. Just one quick question, was TOE-features used on the Intel PRO/1000 MT nics?
April 5th, 2007 at 2:08 am
Toe is not a feature of rhel.
Article ID: 4898
April 17th, 2007 at 4:16 pm
Correct we ran stock RHEL4 U4 drivers.
We are studying performance of RHEL4 U5 and RHEL5 which support Intel’s IOAT and Qlogic iSCSI HBA’s to futher offload the server cpus.
June 11th, 2007 at 6:45 pm
Is my impression correct that LUNs were spread over two 1G Eth NICs in your testing, i.e. two independent channels of 1G Eth each?
IMO, the basic assumption in your test is that the behavior of two independent channels of 1G each will be same as one channel of 2G. You may want to test this assumption, you may be surprised what you find.
July 6th, 2007 at 10:12 pm
I don’t find it at all surprising that two PS3800 arrays with dual controllers would perform at the same level as a single MSA1000. It seems to me that you have presented something rather obvious to those who have worked with both arrays. You might as well have compared sailboats with race horses.