"Analyze the quality of your line, set realistic expectations and then monitor the performance of your line," advises Wayne Lam, vice-president and co-founder of FalconStor a manufacturer of IP networking products. In fact Lam says FalconStor has found that line performance is probably the most significant factor affecting mirroring performance and reliability in a typical mirroring installation.
In the case of a synchronous mirror, where the remote mirror must send a confirmation before a write operation is considered complete, the effects of poor line performance are immediately obvious. "With a synchronous mirror the situation is cut and dried," Lam says. "If you're out of synch you're dead. The worse your transmission line the slower the application." With an asynchronous mirror, where the write is considered successful as soon as the local write is confirmed, the problems can be more subtle, but just as real.
"Typically you do your design not counting on 100 percent performance [from your communication link], but maybe only 75 percent," Lam says. This, along with whatever buffering is included in your architecture, gives a safety margin for the link. However if you don't monitor the line's actual performance, it can slowly degrade to the point where you not only lose your safety margin, but the system gets so far out of synch that the mirror breaks and needs to be what Lam calls 'resilvered' (resnychronized.)
In deciding on the capacity of the link you also need to consider the need to resychronize the mirror if it breaks. This requires more bandwidth than day to day operations and you either need to make sure that you have enough extra capacity in your link or that your service provider can increase the bandwidth of your link almost immediately. (Make sure you define 'almost immediately' in your Quality of Service (QoS) agreement. To a lot of service providers, 'almost immediately', means by next Tuesday.)
If you monitor your line quality constantly you will usually have early warning of timing conditions that might break the mirror. That gives you the opportunity to head them off. There are economic considerations as well. "If you never monitor your line's performance and you're paying for, say, 100 millisecond [latency] and you're getting 200 milliseconds, what a ripoff!"
Lam has another point about keeping your line performance up to the levels in the QoS agreement with your provider. "The squeaky wheel gets the grease is very true in the service world," he says. He advises contacting the service provider as soon as line performance falls below QOS levels and continuing to complain until the problem is fixed.
For more information:Tip: Resource guide to snapshots
Tip: Dos and Don'ts of SAN cabling
Tip: Fibre Channel, part one: Going the distance with SANs
About the author: Rick Cook has been writing about mass storage since the days when the term meant an 80K floppy disk. The computers he learned on used ferrite cores and magnetic drums. For the last twenty years he has been a freelance writer specializing in storage and other computer issues.