How The MacBook Pro Virtual Memory Tests Were Done
Tests that follow were run using Mac OS X 10.6.3 on two models of the MacBook Pro:
- The 2009 17" Apple MacBook Pro Core 2 Duo, 2.93GHz, with 8GB memory.
- The 2010 17" Apple MacBook Pro Core i7, 2.66GHz, with 8GB memory.
The 2010 MacBook Pro is much faster than the prior model, but since virtual memory speed is determined primarily by drive speed, we might expect parity. But it turns out that the prior model also has impaired SATA (drive) performance, and this hurts its performance.
The times displayed in the graphs that follow are the average in seconds of 10 iterations (multiply by 10 for runtime). Small variations of 1-2% percent were seen with multiple trials, usually less than 1% in 64-bit mode; 32-bit mode is a bit more erratic.
A long runtime is necessary because the virtual memory system doesn’t behave exactly the same way for shorter runtimes, especially with the 32-bit kernel.
Results were double-checked for consistency.
The MemoryTester 'vm' command was used for testing.
By default, the MemoryTester 'vm' command allocates 12.5% more virtual memory than physical memory, then proceeds to sequentially access (compare) the allocated memory, which forces the system to swap memory to and from the system drive (page-ins and page-outs).
In essence, the test approach means a fixed workload for all configurations, thus the time taken is a true measure of performance: it is the actual time taken (on average) per iteration. Multiple by 10 for clock time.
Cloning was used to guarantee an identical Mac OS X 10.6.3 system for each test. The Mac OS Finder was quit prior to the test, so that Terminal was the only application visible during the test. This is not required, but was done for consistency each time.
Kernel — 32-bit or 64-bit
Ignorant claims can be found online that a 64-bit kernel is only for those that have gobs of memory. The tests show that a 64-bit kernel is a massive improvement for virtual memory with only 8GB of real memory. The performance tests in the review of Snow Leopard here on this site also show clear benefits.
Running a 32-bit kernel shows very slow and erratic performance with virtual memory. Unless you have crappy software or older hardware that won’t run at all in 64-bit mode, switch to 64-bit mode with Snow Leopard.
Mostly empty drives — best case for HDD
The test used a system install of about 22GB in size. In real life, a drive would contain data, and since hard drives slow down as they fill up, real world hard disk drive results would always be slower.