Brocade & Cisco: Who's Out of Order?

Shares of Brocade Communications Systems Inc. (Nasdaq: BRCD) jumped 20 percent today after a Wall Street analyst upgraded his rating on the company -- but Cisco Systems Inc. (Nasdaq: CSCO) says a critical part of the analyst's rationale for doing so is simply wrong.

John Roy, an analyst with Merrill Lynch & Co. Inc., issued a note this morning raising his rating on Brocade from Neutral to Buy.

In part, he justifies the brighter outlook for Brocade on mounting challenges he believes Cisco will face as it tries to penetrate the Fibre Channel switch market, which is dominated by Brocade and McData Corp. (Nasdaq: MCDTA).

"We believe it will be 12 months before Cisco has the right products to make a run at Brocade and McData," he writes in the note. "Recent information points to the immaturity of the products."

Other factors Roy cites in his note include Brocade's improved competitive positioning against McData; his belief that Brocade has "fixed its inventory problem," which has been a drag on earnings the past few quarters; and strong continued demand for SANs. Based on these trends, Merrill Lynch raised its estimates for Brocade's fiscal year 2003 from $526 million and earnings of $0.03 per share to $532 million and $0.04 per share. For fiscal year 2004, it raised estimates from $631 million and earnings of $0.07 per share to $638 million and $0.12 per share because of better gross and operating margins.The upgrade helped push Brocade's shares up 20.2 percent for the day, closing at $7.26.

However, Cisco heatedly disputes a key portion of Roy's analysis -- namely, his assertion that there are performance problems with Cisco's Fibre Channel switch. Roy writes:

Cisco is convinced that this information -- or misinformation, as the case may be -- about its switches originated with Brocade. "We'll take the high road and assume it's a misunderstanding of our architecture," says Paul Dul, product line manager for the MDS 9000.

In fact, these details about Cisco's switch did come directly from Brocade, as Brocade acknowledges. (Merrill Lynch's Roy could not be reached for additional comment by press time.) But Ron Totah, technical marketing manager at Brocade, says it's not a misunderstanding: He says there are some serious issues Brocade has identified with out-of-order frames -- as well as dropped frames -- in its own testing of the MDS 9509 and 9216.

"On different tests, it seems to exhibit certain problems," Totah says. "It's not surprising -- it's a 1.0 product."For example, he says, using Spirent Communications' SmartBits 48-port full-mesh test, he's seen dropped frames in the ballpark of 0.04 to 0.05 percent on the MDS switches -- well above the error rates on Brocade switches, which he says are roughly one in a trillion. Brocade obtained the Cisco switches, which Totah says are running firmware release 1.1, from IBM Corp. (NYSE: IBM).

These latest allegations come as competitors are also criticizing Cisco's 32-port line card as oversubscribed. Cisco, however, not only owns up to this, but says the 32-port card provides SAN designers more flexible options (see Cisco: Oversubscribed by Design).

But Cisco says there are simply no out-of-order or dropped frame issues with its switch, and that any interpretation that there are is the result of a misunderstanding of its architecture.

First, says Tom Nosella, senior manager of technical marketing in Cisco's Storage Technology Group, it is architecturally impossible for the MDS 9000 switches to put frames out of order within the switch itself. "We never load-balance dynamically across the crossbars -- it's a static mapping," he says.

The only way an MDS 9000 switch could be caused to transmit frames out of order is if there were a topology change introduced to the fabric, at which point, Nosella says, "anybody will have problems with dropped frames... drops are normal and they happen in topology changes on anybody's switch."To handle this eventuality, Brocade, Cisco, and McData switches all include a feature -- called in-order delivery -- that basically freezes the SAN fabric and flushes out the frames that are still in the switch. That eliminates the chance that a frame will end up in the wrong order on an egress port.

Now, according to Nosella, the only difference between Cisco's approach to this issue and its competitors is that Cisco turns off the feature by default, whereas in Brocade and McData fabrics it is typically enabled.

"Because our switch forwards everything in order, [in-order delivery] is off by default," Nosella says. "If a customer is running a very structured SAN design, the chance of having an out-of-order frame is slim to none." In a fully meshed network, he adds, it's advisable to turn on in-order delivery "for safety reasons," in case one of the links in the network goes down.

Thus, according to Cisco, someone would conclude that its switch has a higher out-of-order frame error rate than its competitors only if he or she didn't know -- or deliberately ignored -- that the switch's in-order delivery feature is turned off by default.

Nosella says Miercom, a testing firm in Princeton Junction, N.J., ran the Cisco switch through an exhaustive battery of tests in December 2002, and that it turned up none of the problems that Brocade or the Merrill Lynch note allude to. Brocade's Totah says he's not "not sure why Miercom didn't see" any of the performance problems Brocade has.And what does Miercom say? Michael Hommer, senior manager of lab testing at Miercom who conducted the test on the MDS 9509, says of the alleged out-of-order frames: "This was not anything we noticed in our testing."

Todd Spangler, US Editor, Byte and Switch