The servers were presented with two types of workloads: calculation-intensive and disk I/O-intensive. The calculation-intensive workload presented transactions that repeatedly accessed a small area of the MySQL database, performing only tiny amounts of physical disk I/O. The I/O-intensive workload offered a mix of transactions that spanned an area of the database that was much larger than the server's largest possible size of disk cache memory. "This large database footprint ensured that virtually every transaction would cause a read-from and/or write-to the physical disk drives," according to Nelson's report.
Nelson ran the entire test sequence twice. The first time, both servers were configured with 4GB of main memory. The second time, they were both configured with 16GB of memory.
Performance just a starting point
The testing yielded some interesting results. The Opteron (despite its slightly slower clock speed) delivered higher performance at 100 simulated users: When the servers were loaded with 4GB of memory, the Opteron managed 32,314 TPM (transactions per minute) whereas the Xeon completed 30,406 TPM, a 6.3 percent difference. At 16GB of memory, the Opteron also fared better at 100 users, but the Xeon closed the gap: 30,989 TPM versus 30,667 TPM, a 1 percent difference.
Similarly at the maximum number of simulated users, 500, the Opteron delivered more TPM. When the servers were equipped with 4GB of memory, the Opteron performed 5.2 percent more TPM: 5,081 versus the Xeon's 4,831. At 16GB of memory, the Opteron squeezed out 0.6 percent more TPM: 5,744 compared to 5,711 for the Xeon.
However, when the numbers of users ranged from 250 to 400, the Xeon was the consistent winner, coming out ahead by as much as 3.9 percent more TPM.
These differences in overall raw performance between the two servers are, in my view, fairly unremarkable. The Opteron arguably has a slight edge. That makes the differences in power consumption all the more remarkable. Here, the Opteron was the consistent winner. When the machines were equipped with 4GB of main memory, the Xeon machine burned between 13.1 and 14.4 percent more watts per hour than the Opteron. For example, at 500 users, the Xeon server consumed 209.6 watts per hour, whereas the Opteron consumed 180.8.
The margin widened at 16GB of memory: The Xeon consumed between 20.3 and 21.3 percent more watts per hour. For example, at 500 users, the Opteron consumed 188.2 watts per hour, whereas the Xeon consumed 239.