Our 3.2E 'Prescott' notches its first victory over the Northwood-based P4 and really puts in a fine performance in this test coming awfully close to the performance level of our dual Xeon box. I'm thinking the added L2 cache is helping Prescott here. Enabling HT on this machine results in a 29% increase in requests per second.

The dual 3.2GHz Xeons with 1MB of L3 cache win out in this benchmark by a small margin (as they should). Interestingly enough we see a decrease in performance of about 3.4% when HT is enabled on the Xeons. Are the four logical processors stumbling over each other trying to gain access to cache or the bus? I'm thinking once Nocona is available we'll see far fewer instances of this as the increase to an 800MHz FSB will really help.

I don't think the performance decline is large enough to really motivate you to disable HT on your Xeons powering web servers. You may want to hold on long enough to take a look at database performance before making your decision.

The 3.2GHz 'Northwood' P4 finishes in last place. Enabling HT here gives us another ~28-29% increase in performance. That keeps it directly inline with the increases we saw on our Prescott box.

Finding an interesting MySQL benchmark really turned out to be quite a chore for me. I spent time messing around with several only to be disenchanted with the results. I originally got pretty excited about the Database Opensource Test Suite (DOTS) which required a client and server machine and used JDBC to connect to a performance monitor and the database respectively. I was able to get it working without any problems but for whatever reason I couldn't get the client to put enough stress on the database server to actually end up with reasonable and accurate results. I dropped that and moved onto the Open Source Database Benchmark (OSDB), which still didn't give me what I was looking for.

Eventually I talked to my homie Jason, who's the closest thing I know to a database genius. He suggested I give sql-bench or "MySQL Super Smack" a try as the former is bundled with the MySQL source distribution and is written in perl and the latter is multi-threaded. sql-bench didn't really blow my skirt up either, so I eventually settled on MySQL Super Smack. At the very least, it has a cool name.

Super Smack comes with a couple of "smack" files that can be used to test the database performance of your machine. It will work with Oracle, PostgreSQL and MySQL. In this case of course, we were using MySQL 4.0.17. The first test consists of two straight selects which should give you a reasonable snapshot of the read performance of your database server. If you'd like to take a look at the SELECT test file, here it is. Let's take a look at the numbers from the first test.

It should be noted that I ran the test five times in each instance and averaged the results.

Don't let anyone ever tell you that dual machines don't make kick-ass database servers. Look at how our Xeons knock the dust off the uniprocessor test systems. The Xeons see a ~25.5% increase in performance once we enable HT. 44,502 queries per second is awesome.

The 3.2E 'Prescott' disappoints me here. We know that database servers generally love additional cache and I thought the extra 512KB of L2 cache might swing the results in Prescott's favor. Unfortunately that's not the case. This might have something to do with that deep pipeline. All the same, we see a ~43% increase in performance once we enable HT. You really can't complain about that, can you?

Northwood fared pretty well in this test. It slightly edges the Prescott in both scenarios (HT enabled and disabled) and enjoys a ~39% increase in performance once HT is enabled.

On to the next MySQL test!