Capacity Planning - an Obsolete Notion

There used to be rules and algorithms for sizing systems, storage, and memory. Capacity planning programs would take input on the number of users, expected file transfers, load statistics, and growth estimates, and spit out disk and CPU sizes you'd need.

Forget it. All that stuff is obsolete. In the "E" age, none of this applies.

For example, take Charles Schwab. In early 1998, the brokerage's Web site was handling 6 million hits per day. Then it lowered prices and made a concerted effort to land online accounts, and by last April, its daily Web traffic soared to 75 million hits. The company had to add five mainframes and increase IT staff 50 per cent - to 2000 full-time equivalents - to handle the load.

A little later, it added more mainframes. Capacity planning consisted of buying computers and switches as fast as it could. It also involved working with IBM to find a way to link a bunch of mainframes into the equivalent of a 32-way multiprocessor.

'Spikier' Traffic

The company also found out what you can expect when you commit to e-business in a big way - traffic is much "spikier". Schwab used to design its systems to handle three times its daily peak traffic average. But in its online operations, it had to build capacity to handle three times its hourly peak average, or the same as 10 times its daily average.

Some of the reason for such "spikiness" is self-fulfilling. Make an e-business site easy to use and navigate, make it personalised, and guess what? You'll get a lot more traffic! One online broker told me that more than 90 per cent of all online traffic at places like E-Trade has nothing to with commissioned stock trading - just a lot of investors checking their portfolios many times a day.

For most companies, this will be a triple whammy. You'll be drawing in more site visitors, increasing traffic per visitor, and growing off-business-hour or off-time-zone traffic. But you won't have a clue how that demand will grow over a month, a quarter, or a year.

In a survey of 400 US businesses in the middle of last year, IDC found that the average company expected to nearly double the number of individuals served at peak use times (from 6300 to 12,000) by the end of 1999. They also expected the number of transactions per day to grow from 2000 to 23,000, and the size of their databases in support of their Web sites to grow from 175GB to 1100 GB.

Got that? More than 10 times the number of customers in six months, and more than six times the capacity to support them.

That may be right. But if Schwab is a leading indicator, those metrics are too conservative. Three tips:

Your capacity planning should be based on your capacity to order systems, storage, or memory, and get them installed. You'll be ordering on demand.

Choose an architecture that can scale to at least 20 times what you really think you'll need in six months. (When was the last time anyone chewed you out for having too much spare disk drive capacity?) Don't forget the changing peak-to-average load dynamics, and while you're at it, you might as well plan a 24x7x365x24 (time zones) operation.

In fact, don't plan at all; just do!

John Gantz is Chief Research Officer for IDC Australia

Join the newsletter!


Sign up to gain exclusive access to email subscriptions, event invitations, competitions, giveaways, and much more.

Membership is free, and your security and privacy remain protected. View our privacy policy before signing up.

Error: Please check your email address.

More about Charles SchwabIBM AustraliaIDC AustraliaSchwab

Show Comments