Performance comparison - need advice

Hi,

I’d like to compare two servers configuration :

1 . HP LH3R (4 Xeon 500Mhz, 1Gb Ram, Raid 5 scsi disks)

2 . Dell poweredge 1650 (2 PIII 1.5Ghz, 1Gb Ram, Raid 1 scsi disks)

Which is the best ? More processors with less power or Less processors with more power ?

The application is a web admin with 100,000 documents databases and full-text indexes.

(i tried to register NotesBench.org, but never received registration emails)

Thank’s for advice

Mathias

Subject: performance comparison - need advice

It’s going to be difficult to give accurate feedback without more server specifications and details about your application(s). I’ll do my best though, making a few assumptions:

Drives - Assuming the drives are identical, the Raid 1 is going to be noticeably faster on reads and much, much faster on writes than your Raid 5. If the servers have enough disks, though, you’ll be best off going with a configuration of: 1 disk for OS/Domino, 1 disk for view index building, 1 array (preferably Raid 10) for data storage, 1 disk for swap file, and 1 disk for transaction log (if you use it). If nothing else, at least try to dedicate a drive to view index building, separate from your data drive(s).

RAM - The more RAM the better, though your load will determine if you really need more than 1Gb

CPU - Generally speaking, more CPUs are going to help you under load, while faster CPUs are going to make some things (like HTTP processing time, agents, etc.) seem faster. With a faster CPU, view indexing will run quicker, though with 4 CPUs you could run a couple of update tasks, possibly updating multiple indices faster over time. The front-side-bus of the P-III server is probably going to be higher than that of the Xeon server, which would increase your memory throughput. Bottom line: your load will play a big role here (in particular, the maximum number of HTTP threads you need to run). It is probably the case that your P-III server’s cpus/motherboard are so much quicker than your Xeon that it more than makes up for the extra context-switching involved with having half the number of processors.

Like I said, it’s hard to say much without concrete testing, but hopefully this helps you some.

Subject: Nitpick

In other cases, I could understand your position on the CPU count vs. MhZ issue. But in this case, the speed difference is profound between the options: 4 Xeon 500Mhz vs. 2 PIII 1.5Ghz. While dealing with the smaller number of CPUs means more context switching on the PIIIs, the clock cycle difference is considerable. The difference is also emphasized by the fact that we’re comparing 2 CPUs to 4, rather than, say 1 to 2. For Domino, a second CPU scales pretty close to linearly, but the 3rd and 4th see diminishing returns pretty quickly according to NotesBench test, particularly on the NT platform.

The paralleling requirements would have to be high indeed to make up for the 50% gain in available clock cycles on the PIII pair. Given the losses experienced in a 4 CPU configuration, I cannot imagine a context in which 4 500Mhz CPUs would outperform 2 1500Mhz processors. Perhaps if there were considerable differences in cache sizes, but as you observe, FSB differences also play in favor of the dual config.

By the way, check out the NotesBench consortium’s reports for more real-world detail.

http://www.notesbench.org

Subject: RE: performance comparison - need advice

Thanks, it helps a lot.The first structure of the main DB had 200 views ! The DB I worked on never have more than 20 views, and I implemented full text indexes.

No transactionnal logging at all on the DBs.

I have lots of scheduled agents, an http requests.

Subject: RE: performance comparison - need advice

Where can I find the documentation on how to “dedicate a drive to view index building”?

Thank You,

Joni Snyder

American University

Subject: RE: performance comparison - need advice

I think it’s in the administrator help. If you’re using Win32, though, you can simply rely on the fact that Domino uses the path in the TEMP environment variable. Just change that on the machine to point to the drive and you should be good-to-go. As Domino uses it you should see a folder appear called “E.Data” or something similar.