The behind-the-scenes march by storage system vendors to switch disks within the array began a couple of years...
By submitting your email address, you agree to receive emails regarding relevant topic offers from TechTarget and its partners. You can withdraw your consent at any time. Contact TechTarget at 275 Grove Street, Newton, MA.
ago, but recent announcements from providers of storage silicon suggest that the pace is quickening. As a result, storage buyers should begin to see high-end arrays with better availability and performance characteristics than what they are currently used to.
This spring, PMC-Sierra announced a 4 GB cut-through switch, or loop switch, which the company bills as a big improvement to the intelligent port bypass circuit (PBC) it announced in 2003. More recently, the semi-conductor company Broadcom, which has shown an intense interest in the storage market, announced an entire 4 GBps 16-port fabric -- as opposed to loop -- switch designed to go inside high-density arrays.
The trend of switching disks within the array was arguably started by Vixel when it introduced its InSpeed switch on a chip (SOC) more than two years ago. Vixel has since been acquired by Emulex, which recently reported it had shipped more than 2 million InSpeed ports across at least 15 storage systems, says Brian Reed, Emulex vice president of business development. InSpeed customers include vendors such as BlueArc, Hewlett-Packard and Network Appliance.
What does switching fix? According to Peter Wong, manager of strategic marketing at PMC-Sierra, building a Fibre Channel array out of disks and PBCs on a Fibre Channel Arbitrated Loop (FC-AL) loop introduces "a myriad of problems." Those problems include maintaining signal integrity and achieving consistent performance. But PBCs biggest drawback is arguably its poor "diagnosability," or the difficulty one has in identifying a failed drive if it's on a PBC-controlled loop.
A loop switch, in contrast, maintains the simplicity of FC-AL, but allows "the user to dive down and tell what drive is having trouble," Wong says. At the same time, a loop switch can improve system performance through reduced arbitration -- in other words, by reducing the number of hops (PBCs) you need to go through in order to get to a drive.
Broadcom's fabric BCM8441 takes the concept of switched disks one step further. Because it is a fabric switch, it frees array manufacturers from FC-AL's limited device support of 126 devices per loop. In contrast, a Fibre Channel fabric supports more than 16 million devices. That's important for high-end arrays, especially when you consider the small form factor drives joining the market.
In the past, "126 devices wasn't an issue," says Michael McDonald, senior director of storage area networking for Broadcom, but with small form factor drives, "we'll be getting up to 126 devices fairly quickly."
Furthermore, says Arun Taneja, analyst and founder of Taneja Group, a 4 GB fabric switch can support disk drives at both 1 GB, 2 GB and 4 GB; in a loop, bandwidth is capped at the slowest device on the loop. That may be important for modular arrays, where customers buy a shelf of 2 GBps drives today, but 4 GBps drives tomorrow.
Nor surprisingly, existing loop switch vendors aren't sold on the concept of a fabric switch in the array. Both PMC-Sierra's Wong and Emulex's Reed point to additional cost and complexity of fabric switching as prohibitive and unnecessary. "Our customers report that [fabric switching] is a cumbersome solution, and overkill for what more customers are looking for," says Wong.
But if Broadcom's BCM8441 finds favor among array manufacturers -- and according to Taneja, it is currently being evaluated by a large number of them -- fabric switching's complexity should not be passed down to the storage buyer. "Whatever minimal level of complexity a fabric brings will be completely transparent to the end user," he says, "and by that I mean the IT administrator."