iSCSI Takes on Fibre Channel

Fibre Channel cant kick sand in the face of iSCSI anymore, according to eWEEK Labs testing.

iSCSI has been on the storage networking scene for several years now, but it has lagged behind Fibre Channel in several key enterprise areas, including performance, reliability and maturity. iSCSI-based products also have been relatively few and far between since the standard was ratified in 2003.

Click here to read eWEEK Labs take on the emerging iSCSI market.

However, iSCSIs performance in head-to-head tests against Fibre Channel, as well as greater availability of iSCSI-based products, shows that iSCSI has matured into a viable substitute for Fibre Channel when storing data for typical business applications such as e-mail and workgroup databases. Its a good storage bet overall for SMBs (small and midsize businesses), and larger companies can use the technology for cost-effectively connecting low-end and midrange servers to SANs (storage area networks).

When compared with Fibre Channel, iSCSIs core strengths are ease of implementation and cost efficiency. Fibre Channel, on the other hand, has speed and maturity on its side—important considerations when talking about moving mission-critical data on storage networks.

iSCSI does consume server processor cycles, although storage managers have gotten around this through the use of TCP Offload Engine, or TOE, cards. For enterprise applications priced per CPU, however, we recommend sticking with Fibre Channel.

Interoperability—or lack thereof—was a big impediment to widespread enterprise use of iSCSI. iSCSI is a standard, but with no iSCSI initiator software built into major operating systems, iSCSI vendors were left to their own, proprietary devices. Now, however, most major operating systems include iSCSI initiator software.

iSCSI initiator software has been either included with, or freely downloadable for, AIX, Windows, NetWare and HP-UX for a few years now. Last year, Linux and Solaris followed suit. (The notable exception is Mac OS X.)

While there have been good reasons for the past couple of years to think iSCSI when evaluating SAN solutions, performance has been a big sticking point. All else being equal, a solution that cant get data where it needs to go in a timely fashion isnt a solution at all.

E-mail is an important market for iSCSI, and it is also an application whose performance depends on high-quality storage. For this reason, we chose to focus our testing on detecting performance differences between iSCSI and Fibre Channel storage systems when hosting e-mail data.

Read more here about Labs iSCSI/Fibre Channel test methodology.

During tests using Microsofts Exchange Server 2003 Enterprise Edition, we found the performance gap between Fibre Channel and iSCSI to be negligible over a range of workloads. For very large data loads, Fibre Channel is still the way to go, but our tests show that iSCSI will be adequate for most applications.

We ran benchmarks using Microsofts LoadSim 2003. We configured our storage units and our SUT (server under test) for production-level settings.

We ran each test iteration twice, storing the Exchange mail stores first on a Fibre Channel-based Dell/EMC AX100 system and then using an iSCSI-based Dell/EMC AX100 storage system.

LoadSim 2003 populates Active Directory with user accounts and creates mailboxes for these accounts, complete with messages and distribution lists. We used an MMB3 profile—the workload profile that vendors and analysts use when publishing LoadSim results—that ran each user account through the typical activities seen during an 8-hour business day. For example, the LoadSim 2003 server orders virtual users to create, delete and modify messages and appointments in their accounts.

The primary metric for gauging performance in LoadSim 2003 is a latency measurement for the 95th percentile of transactions. The smaller the score, the faster an operation is completed.

In our first round of tests, running with 500 virtual users, the Exchange server turned in a score of 279 when using Fibre Channel storage. (That is, it took 279 milliseconds on average to complete the top 95 percent of transactions.) Running the same 500-user load with iSCSI, the LoadSim 2003 score went up only slightly, to 299. (Remember, the lower number is better with these tests.)

iSCSI was slower than Fibre Channel each time we increased the user workload, but the gap was always minuscule.

Interestingly, both the iSCSI and Fibre Channel storage units hit their breaking points at 1,200 virtual users. At this point, the Fibre Channel LoadSim 2003 score was 932, and the iSCSI score was 954.

Microsoft considers only scores under 1,000 ms as acceptable, and neither storage system garnered a passing score while servicing more than 1,200 users.

Next Page: Cost factors.