IT Troubleshooter | Harper Mann » Are SANs really as infallible as vendors would have you believe?

June 19, 2006 | Comments: (0)

Are SANs really as infallible as vendors would have you believe?

TAGS: Networking

The original rational behind the enterprise SAN and fiber channel fabric market was to put in place a world-class infrastructure for protecting data. By dissolving the direct-attached relationship between servers and storage, the idea was that when servers failed, the storage would still be accessible.

But studies about downtime causes are steadily surfacing data that suggests that SANs themselves go on the fritz relatively frequently. This recent vendor-produced white paper ("Why Email Fails") by MessageOne, for example, cites SAN failures as one of the leading overall causes of messaging system failures, and suggests that compared to other causes of failures, SAN outages tend to lead to a considerable amount of down time.

So what are some of the common afflictions in SAN technology? Jon Toigo, principal of Toigo Partners recently enlightened me:

"There are a number of problems with SAN technology that are giving enterprise IT pros problems.

One is the poor set of standards upon which they are based. You could fill a whole library with fiber channel standards. They've been developed primarily by players in the industry. And the really interesting thing about their standards is that you can build a switch to the letter of the standard with absolute certainty that the switch will not work with competitor switch, which is also built to the letter of the standard. Now, you figure it out. When is a standard not a standard? When has the storage standards process been hijacked by the vendor community? I once referred to a fiber channel standards group as the Taliban of the storage industry, and I got in trouble for it. Somebody responded that the comment was over the top, but someone else noted that it was also a little unfair to the Taliban.

Secondly, SAN technology is prone to a lot of human factors in the configuration and set-up ... failures that are based on the fact that people may not understand how to create a zone or they may not understand how to properly configure their SAN. There are lots of configurations and complexities that require ongoing fine-tuning and optimization, and create a lot of dependencies. Which, of course, means lots of little things that can break.

The other common criticism of SANs is that they are extremely difficult to manage. If you have a heterogeneous environment -- meaning you've got disk arrays from multiple vendors -- it is almost impossible to manage a SAN. And there have been ongoing efforts, like SMI, at the storage networking industry association, to try to come up with the universal holy grail of management, except that it's all held hostage by the vendors. To a vendor, making management easy means that you also make it easy to deploy his competitor's product in the same storage fabric as his, and he doesn't like that idea. So the storage vendors' participation in management standards has been halting at best. For end users, it's tough to see when a SAN problem is building, and it's really tough to quickly troubleshoot and rectify the problem when it occurs. So the only way around it, to get a little more efficiency and a little more resiliency out of the SAN, is to buy all the pieces from a single vendor. So while the price of compute has dropped (on a per gigabyte basis) at the rate of 50% per year since the mid- 90's ... the price of a storage array made up of a bunch of commodity disks has actually accelerated at a rate of 125% per year."

I completely agree with Jon's points. SAN vendors have sold a "bill of goods." SAN complexity is through the roof ... possibly two orders of magnitude over conventional storage, which makes them nearly impossible to manage. Most failures are due to configuration changes, not hardware or software bugs or breakage. Add that to the extra virtual file system layer needed to manage SAN requests -- which is on the order of a whole other OS with the concomitant problems with races and resources -- and you've got even more trouble.

The big SAN vendors want it complex so they can make the case for how expensive it all is. If you look at the per disk cost for SAN approved disks, they are on the order of $3000 or $4000 each, plus the extra charges for going > 1TB and similar. As an IT director, I don't see any value in paying more just because I went over some arbitrary value like 1TB. It's also galling to pay thousands of dollars for the same damn disk that cost $150 bought raw from the manufacturer.

Posted by Harper Mann on June 19, 2006 07:39 AM

RATE THIS ARTICLE: