The iSCSI-Fibre Channel horserace
- 10 November, 2004 10:17
The idea of comparing the iSCSI and FC (Fibre Channel) transport protocols as if they were opposing thoroughbreds at the Kentucky Derby may sound childish, but it's a temptation often difficult to resist.
There is a good reason to line them up head-to-head: These two technologies can be deployed as alternative or complementary transport for a storage network. In fact, if you start building your makeshift SAN today, you can be equally successful with FC or iSCSI -- although considerations such as cost, performance, and maturity of the products can sway your choice.
Although the other two factors leave room for give and take, performance is usually an inflexible requirement. The amount of data traffic that will flow on your future SAN is obviously what makes the pendulum of your decision swing one way or the other.
Which means that, other factors notwithstanding, a need for top performance will translate into selecting FC over iSCSI. It's no secret that iSCSI has been stuck at a rate of 1Gbps, while 2Gbps FC transport is already quite popular.
Furthermore, products based on FC transport are achieving even higher speeds. Earlier this year, QLogic deployed the SANbox 5200 switch, which combines 2Gbps connectivity to hosts and storage devices with 10Gbps inter-switch links. This was, I believe, the first practical implementation of that higher speed on FC.
As you probably know, 10Gbps FC is not backward-compatible with previous implementations, which cools much of the deployment enthusiasm for new devices, and it also provides a rationale for the architecture of the QLogic 5200.
Luckily, lower-speed FC transport does not trigger compatibility problems. In fact, researchers realized that they can maintain backward compatibility even if they double the FC clock twice, at 4Gbps first and then at 8Gbps.
Not surprisingly, these new, faster FC devices are already entering the market. In fact, at the recent Storage Networking World (SNW) show, Brocade announced a 4Gbps-capable fabric switch, the SilkWorm 4100, which comes with 16, 24, or 32 backward-compatible ports.
At the same show, Emulex presented its 4Gbps novelties, including the LightPulse LP11000 HBA and the Helios I/O Controller.
Keeping in mind that earlier this year Emulex announced 4Gbps support for its SOC (switch on a chip), an embedded device that mounts on storage arrays, it's reasonable to speculate that some OEM may soon come up with storage targets capable of 4Gbps transport, thereby completing the SAN spectrum for that speed. Other switch and HBA vendors will probably follow.
If you are wondering why I spent most of this column on FC, the reason is that not much is happening to increase the speed of iSCSI transport (but I'll be glad to be corrected on this if you've heard differently).
Nevertheless, at least one vendor, Chelsio Communications, has come up with a full-size, PCI-X iSCSI HBA, capable of 10Gbps. This speed demon, dubbed the T110 10Gbps Protocol Engine, was also announced at SNW, and its specs are as imposing as its long name.
You can read all the details at the Chelsio Web site, but according to Chelsio, the T110 can move data between two iSCSI points at the heartbeat-hastening pace of more than 800MBps and can handle more than 500,000 IOPS (input-output operations per second).
On the iSCSI stage, the Chelsio T110 can play both initiator and target roles, which opens the possibility of SANs transporting data faster than the traditional 1Gbps GbE and also faster than FC.
Obviously, now it's up to the SAN vendors (some, such as EqualLogic, have already shown interest in the T110) to make that possibility real, finally removing the performance handicap from iSCSI SANs and make that horse race more interesting. Until we hear more about 10Gbps iSCSI, my money stays on FC transport.
Mario Apicella is a senior analyst at the InfoWorld Test Center.