Some directory lookup benchmarking figures that rock

As promised last time, I want to share with you some eye-opening numbers that HP's Wook Lee presented at last week's Directory Experts Conference.

Working for HP (even as a directory architect) does have hardware benefits. Lee gets to experiment in his lab with all the new PC and server hardware just as soon as they're available. He and his partners decided to try benchmarking new processors using directory lookups. The five machines tested were configured as:

1) Dual 2.2 GHz x64 Opteron processors, 14G-byte RAM.
2) Quad 2.2 GHz x64 Opteron processors, 14G-byte RAM.
3) Dual 1.5 GHz IA64 Itanium II processors, 16G-byte RAM.
4) Dual 733 MHz IA64 Itanium I processors, 9G-byte RAM.
5) Dual 2.4 GHz x386 Xeon processors, 4G-byte RAM.

To continue reading, register here and become an Insider. You'll get free access to premium content from CIO, Computerworld, CSO, InfoWorld, and Network World. See more Insider content or sign in.

As promised last time, I want to share with you some eye-opening numbers that HP's Wook Lee presented at last week's Directory Experts Conference.

Working for HP (even as a directory architect) does have hardware benefits. Lee gets to experiment in his lab with all the new PC and server hardware just as soon as they're available. He and his partners decided to try benchmarking new processors using directory lookups. The five machines tested were configured as:

1) Dual 2.2 GHz x64 Opteron processors, 14G-byte RAM.
2) Quad 2.2 GHz x64 Opteron processors, 14G-byte RAM.
3) Dual 1.5 GHz IA64 Itanium II processors, 16G-byte RAM.
4) Dual 733 MHz IA64 Itanium I processors, 9G-byte RAM.
5) Dual 2.4 GHz x386 Xeon processors, 4G-byte RAM.

They first repeatedly queried on port 3268 from forest root for (instanceType=-1). Since there is no instancetype of (-1), this insured that every row would be touched but no data would be returned. After a number of iterations, they were assured the cache was full (so that no disk access would skew the results) and re-ran the test. The Xeon-equipped machine simply couldn't handle the load and timed out before finishing. For the others, the x64s were almost twice as fast as the IA64s. Interestingly, the quad-processor x64 ended up about 5% slower than the dual processor model. Lee thinks there may be some tuning issues yet to be determined in the quad model.

Lee and his partners then launched another test with a two-part query: (employeeID=*) and (InstanceType = -1). This limited the query to the 213,336 entries that had employee numbers so that even the Xeon machine could compete. Eliminating the quad x64 machine left the remaining four with the following "entries per second" returned:

* 25,941 for the dual 2.2-GHz Opteron. 
* 15,677 for the dual 1.5-GHz Itanium II.
* 3,482 for the dual 733 Mhz Itanium I.
* 3,560 for the dual 2.4 GHz Xeon.

That's right, the x64 was more than one and a half times faster that the Itanium II and a whopping seven and a half times faster than the Itanium I! For database work - and, after all, a directory is simply a special case of a database - these new machines are really going to rock. Start trying to get them into your budget just as soon as possible.

Read more about software in Network World's Software section.