[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: LDAP load testing (Was: Re: slapd lightweight dispatcher)



Rick Jones wrote:
Howard Chu wrote:
Rick Jones wrote:

Indeed, using something _considerably_ newer than RHEL 3 would be goodness :) My guess about the memory usage would be that Linux on Itanium is 64-bit only, so the compilation of the bits would be 64-bit.


Yes, that was the first thing I checked, but in fact sizeof(int) still == 4 on this box.

sizeof(int) is 4 on all 64-bit platforms - unless they are ILP64. Linux on Itanium is LP64 which means longs and pointers are 8 bytes. I'm pretty sure that "all" 64-bit OSes with the exception of Windows 64 are LP64. Windows 64-bit is P64.


I guess my previous comment wasn't clear enough - I said a 32 bit server would essentially max out at 1 million entries. Our tests were all on pure 64 bit code, since we were aiming for 10 million entries to match the results posted by Neil Wilson. Unfortunately we didn't have enough RAM in our machines to fully duplicate those tests.

Without sufficient RAM, those tests all devolve to simply measuring the speed of the disk subsystem, which really isn't what we're interested in.

Also, there are at least two "flavors" of rx1620 - the one that comes with 1.3 GHz CPUs and the one that comes with 1.6 GHz CPUs. The URL above doesn't seem to say which it was. The 1.6 GHz verison has a faster FSB than the 1.3.


This machine has a pair of 1.6GHz CPUs; I don't recall at the moment if they are 3MB cache or 6MB cache. We'll update the blog entry with the details.

I don't think the rx1620 supports the 6MB cache 1.6 GHz CPUs. http://www.hp.com/go/rx1620 would probably show what is supported.

Note - it's a 2620, not a 1620, and the spec page shows that 1.6 Ghz CPUs are available in either 3 or 6MB cache configurations, all at the same FSB speed.
http://www.hp.com/products1/servers/integrity/entry_level/rx2620/index.html


Similarly, the T2000 comes (IIRC) in more than one flavor - they can vary in frequency and number of cores.


This has the identical CPU as that used in Neil Wilson's test - 1.0GHz, 8 cores.

It would be very interesting to see if this benchmark gains anything with 1.2 GHz cores. I'd be interested to hear more about how the benchmark works - offline email may be best for that. I have a amttering of "interesting" hardware at my disposal.

Agreed, we can take the rest of this offline.

--
 -- Howard Chu
 Chief Architect, Symas Corp.  http://www.symas.com
 Director, Highland Sun        http://highlandsun.com/hyc
 OpenLDAP Core Team            http://www.openldap.org/project/