Alex Electronics

I benchmarked Apache 2 on one server with siege running on another. I used three different benchmarking files - one is plaintext, one is a short PHP script with a loop of 1,000 iterations, and one is a PHP script with two nested loops of 1,000 iterations. Different httpd.conf options were tested with the plain text file. Caching was off for the first test and on for the second and third while Keep Alive was on for the first two tests and off for the third. Caching and Keep Alive were on for both of the PHP tests. Siege was configured to run three tests with 32, 64, and 100 concurrent users over a period of 10 minutes. I used two T2000 servers connected with one Fast Ethernet cable from each to a Cisco switch. I did not have a Gigabit switch or a crossover cable to attempt a faster link. The test data is located here. Read the rest of this entry »

I have been very busy with other non-server related activities, so I have not had a chance to post. The second server has arrived and both servers now have GCC 3.4.6, Apache 2.2.0, PHP 5.1.2, MySQL 5.0.19, and Perl 5.8.6 running correctly. I installed Siege 2.64 from JoeDog <http://www.joedog.org/siege/index.php>. Benchmarking with a plaintext file has a transaction time of 0.01 - 0.08 seconds, no where near the server’s limit. Transaction times greater than three seconds appear throughout the test. I think this is due to running out of availiable ports on one of the client box. Setting keep-alive on the server side does little to combat this problem. Otherwise, 500 transactions per second is a very good number. Since the plain text files are nowhere near the server’s capabilities, I used a small PHP script with two nested for loops of 1000 iterations. It takes about five seconds to get the whole page from the server. This page gives a better representation of the server’s capability. With this page, the transaction rate drops with increasing numbers of concurrent users. The transaction rate rose with the number of concurrent users with the plain text benchmark because the server was nowhere near full load. One possible problem with the loop script is bandwidth - every iteration of each loop outputs a number, consuming bandwidth like mad. The script just about maxes it out. I may have to modify the script to use less bandwidth or use a gigabit switch to make the test less network-dependant. The system currently uses a Cisco Fast Ethernet Workgroup Switch.

I also ran a series of PHP benchmarks from PHP Lens <http://phplens.com/benchmark_suite/>. The results of both the siege and PHP benchmarks should be online soon.

The second server arrived today. I will get it set up in a couple of days. I am in the process of compiling Apache 2 and PHP on babylon. The hostname of the first server is babylon.  I’m not quite sure what I’m going to call the second server yet.

First, two server power cables appeared on my front porch. Apparently, Sun sent a second server. Possibly because I submitted the form twice, but I only submitted it a second time because I did not get a reply in over a week. Oh, well! I might get better benchmarking results using a second server and a Gigabit switch so I can run the benchmarking script at full speed.

Second, I finally figured out that DBD::mysql does not like 64 bit integers. After another Perl recompilation, the DBD::mysql module finally installed without any significant problems. The first run of the benchmarking scripts looks ok. After a little tuning in /etc/my.cnf, things are looking a little better. With the key buffer set to 512M, the benchmark script never takes it over 16M. Hmmm, I may need to adjust the script to better test the server’s capabilities. I will post benchmarking results after I have run a few more tests.

Trouble! MySQL will not compile with gcc 4. There is an error with symbol isinf. Apparently gcc 4 processes symbols differently than gcc 3. I could not find a workaround. Time to build gcc 3.4.6, the latest version of gcc 3.

Three hours later…

Now Perl 5.8.8 will not compile. This is one of the gripes I have about Unix-based operating systems: there are odd compatability problems associated with certain applications, especially compilers. Workarounds are not easy to find. Perl 5.8.6 compiled successfully with gcc 3.4.6. Let’s see if this fixes the sql-bench scripts…

Installed binutils and compiled gcc 4.0.2. sql-bench perl scripts segfault, so recompiling perl goes on my to-do list. Since MySQL was compiled with the native Sun ld and as, recompiling MySQL also goes on my to-do list. The gcc build took about 4 hours. I built it with multithreading enabled, so hopefully future builds will be faster.

Built and installed MySQL 5.0.19.  The build took about three hours.  Installed a /etc/init.d script to start mysqld.

I installed Apache 2 and PHP from www.sunfreeware.com.  After many reboots and config file edits, I finally managed to get httpd (Apache 2) and PHP working together.  At least I did not need to complile them!  I might do that later to optimize the software for the platform.

A mammoth box arrived on my door step today. A mammoth box with Sun packing tape. I wasted no time in lugging the thing up the stairs and opening it up. I had purchased a couple of DB-9 to RJ-45 adaptors from Fry’s over the weekend, and had assembled what I thought to be the correct adaptor for the RS-232 connection. Upon connecting the server to my computer, I realized what the “crossing over” the manual talked about actually meant. I busied myself re-assembling the adaptor and plugged it in - success! I configured the login and typed poweron. The server roared to life. Most servers have noisy fans by nature. The T2000’s power supply fans start spinning as soon as the cables are connected. Bad sign in terms of fan noise. The two one-inch power supply fans are nothing compared to the three three-inch main fans. They produce a very loud roar on start and will blow papers across a desk.

I popped open the cover a little later to see what was under the hood. The server is intricately constructed - the IDE and power cable from the power supply/DVD drive IDE board is made up of a bundle of tiny wires. There are what appear to be flat coil inductors on small cards attached to the motherboard. The motherboard is even constructed in two sections with a thin piece of flexible board connecting them. All of the RAM modules are labelled individually. There is also an air duct on the inside of the cover to direct air from the three wind tunnel fans across the processor heat sink and fans. The four serial ATA cables are carefully bent and individually labelled. The hard drives are also 2.5″ Fujitsu MAV2073RC SAS drives - which previously I had not known existed. One thing I also figured out: Sun sells replacement hard drives for about $500, while the bare drive can be had for $700 elsewhere. Wow. Good for Sun! If I need more, there’s no question where I will go: Sun. They must have some sort of quantity discount.

That one server makes my office sound like a datacenter. I will have to be very carful about putting papers near it because of the air currents it generates. This thing sounds like a powerful server when it is running. I can’t wait to start benchmarking!

I recieved an email yesterday (day 8, not counting weekends) with the following subject: “Congratulations and Welcome To the Sun Fire T2000 Try and Buy Program!”  The body of the message said that the server would be arriving in less than two weeks.  Nice!  The email also said that the trial period is 60 days from shipment from Sun (making this day 2 of the trial), so hopefully it will get here soon.