According to Marc Staimer, president of Dragon Slayer Consulting, a speaker at last week's Storage Decisions 2001 conference, users must first decide which level of availability they can live with. The best way to determine this level, he said, depends on how two key questions are answered: How much do you have to spend to get to five 9s? And, what happens if you don't get there?
When it comes to the potential cost of getting to five 9s, the old 80/20 rule applies. "The first 80% [of the efforts to make a system highly available] will be 20% of your cost," Staimer said. "The last 20% will account for 80% of your cost."
The key is to identify the crossover point at which the effort to achieve high availability becomes excessive. "You don't want to spend more than what your value is going to be," he said.
Staimer cautioned that users also need to consider the flip-side of availability cost: What happens if five 9s cannot be achieved? Here, users need to evaluate the business cost of potential downtime events in terms of lost revenue. There are also a few less easily translated costs, such as the potential loss to employee productivity in the event of downtime, he said. "You might have lots of people sitting around, twiddling their thumbs, drinking coffee and
If users decide that achieving five 9s is in fact what they need for critical applications, storage area networks (SANs) can play a key role, according to Staimer. "SANs have become the critical path in a lot of five 9s environments," he said.
Users shouldn't be fooled, however, when it comes to five 9s. Within the term 'five 9s,' there are still different levels of five 9s available. This is especially true in the case of SANs, where five 9s can mean zero loss of SAN capabilities (something akin to RAID 1, Staimer said), or it can mean reduced SAN capabilities (equivalent to RAID 5).
Staimer's example of five 9s achieved with full SAN capabilities includes full bandwidth available all the time. Five 9s with reduced SAN capabilities implies reduced bandwidth and a higher probability of path congestion.
For a SAN environment that requires five 9s with full capabilities, Staimer recommended the use of director-class switches." Directors allow you to engineer environments so you have less inter-switch links," he said. Directors are also less likely to cause bandwidth to become oversubscribed.
For a SAN environment that requires five 9s with reduced capabilities, Staimer recommended the use of a multiple switch environment, which he also called "core/edge networking."
Staimer cautioned about hidden, back-end costs in core/edge networked switch environments, however. While he noted that core/edge switches are less costly than director-class switches, Staimer said that they won't mask failures, so it's very easy to become over-subscribed. "As you add switches, you have to add ports," he said.
In your own evaluation of the level of five 9s required, Staimer recommended that you "mix and match" both core/edge and director-class switches in your SAN environments to best suit the availability needs of your applications and systems. "The key," he said, "is determining where to implement, with what and when."View Webcast associated with this session Storage Decisions 2001 Special Events Page Achieving perfection in uptime