In case you’ve been checking my pictures site you may have noticed that the results for the MySQL benchmarks are available for some days now. It’s just that I didn’t find the time to write the posting to glue them togehter until now.
The Test
Since MySQL is presumably the most used database for web applications, I found a test of MySQL’s performance on the Sun would be just due dilligence. As there’s a lot of benchmarks availbale, the first obstacle to overcome was to pick the right one. I opted for Super Smack 1.3 for some reasons. First, it comes with a set of predefined tests. So it allows you to run the same tests I ran and compare the results. Second, Super Smack is able to spawn a lot of threads to simulate multiple clients. A clear adavantage over the benchmark shipped with MySQL which is single-threaded.
This time, I didn’t included measurements for my trusty Proliant DL380 for the machine was needed in another project. So it was a one on one race between the Lynx Workgroup Server 4300 and my Sun Fire T2000. I ran tests for 1, 10, 20, 30, 40, 50, 60, 70, 80, 90 and 100 simulated clients, each of them requesting 1000, 10000, 100000 and finally 1000000 transactions. The first set of tests consisted of just ‘SELECT’ statements whereas the second set was a mixture of ‘SELECT’ and ‘UPDATE’ operations. The test were run locally on the machines. I tried to run them over the LAN as well but gave up after realizing that the traffic generated by Super Smack easily saturated the cable, rendering the whole network unusable.
The Results
As you can see, the Lynx WGS 4300 and the Sun have a head-to-head race in almost all tests where only SELECTs are used. But you may also note that sometime the Lynx drops in performance considerably. That’s when the MySQL server on the machine wasn’t able to respond in time. Mind you, it didn’t crash. It just didn’t respond to requests anymore because of a backlog that became way too big.
You can also see that there’s a problem with the Lynx when you take a closer look at the timings. You’ll note that the Sun is a lot more responsive to requests, almost never taking more than 2 milliseconds to respond. In contrast, the Lynx got saturated quite easily and sometimes took a whooping 16000 milliseconds to respond. It also dropped some connections completely because too many requests had lined up in the server’s backlog. Clearly no user would wait for 16 seconds when he has sent off an request but instead would consider the application to have crashed.
The picture changes if there’s a lot of UPDATEs to be done. The Lynx offers a lot more performance than the Sun.
What again impressed me was the steady behavior of the Sun. No matter whether under light, medium or heavy load, the Coolthreads just prodded on, uncaring. Although the number of transaction reached is not spectacular, it’s enough to support even a large site with lots of requests. It’s also worth to be noted that the MySQL server used on the Sun was a 32 bit executable (since that’s what you get from Sun and from Blastwave) whereas on the Lynx a 64 bit executable was used.
Actually, that’s the only point where I hold a slight grudge against Sun. I’d have expected them to provide not only 64 bit versions of the Open Source Software provided but also current versions of the tools in the Free Software bundle. Neither is the case. Since the latest MySQL version for Sparch downloaded from MySQL.com refused to work and I had to be back on the road again, I can’t say whether using MySQL 5 and it’s conmpressed protocol would have made a difference. I’ll give that another try once I’m back in the office but can’t guarantee I’ll be able to finish the test runs. They’re running for three to four days and I currently just don’t have the time.
Follow these links and you’ll find the results in the formats for Excel and Spreadsheet applications capable of reading the Open Document Format.
[ All posts about my experiences with the SunFire T2000 >>> ]
