Saturday, March 31, 2007

Windows ReadyBoost

On Wednesday I purchased a 1GB USB stick for $20 from Dell. I think it was part of a promotion Dell had - in any case, I needed a USB stick, and this one was compatible with "Windows ReadyBoost", a new feature in Windows Vista, which I conceivably could use.

There is debate as to whether using ReadyBoost is even useful for a decently fast computer. All windows is doing is using the USB stick as cache. If some other form of cache is faster, more abundant, or if lower levels of cache aren't even being filled, ReadyBoost is useless indeed (assuming windows uses it only as cache). So, I decided to run some tests on my own computer to see if using ReadyBoost speeds up, slows down, or doesn't affect my computer's performance.

Test: 3DMark05

With ReadyBoost:
Test1: 4545
Test2:4552
Test6: 4521

Without ReadyBoost:
Test3:4558
Test4:4559
Test5:4548

n.b. I rebooted after test 4.

Using a program called Disk Write and Read Benchmark written by Roy Longbottom, I produced the following graphs:


Using ReadyBoost:

NOT using ReadyBoost:
In this case, higher bars are better. The files named NRB are the Not Ready Boost tests. As you can see, there's obviously a slow down when using ReadyBoost.

I also ran a program called BusRead, whose description is given by the author:
"*These measures performance on transferring data from memory. Ten tests
* are run with data size 4, 8, 16, 32 etc. Kbytes with a default maximum
* of 50% of main RAM size for 128 MB or greater. Above 128 MB options are
* available to select maximum of RAM size minus 64 MB. Default running
* time is a minimum of 100 milliseconds per test. Extended sequences of
* assembly code instructions are used and multiple registers, attempting
* to transfer data as fast as possible. Results are calculated in millions
* of bytes per second. Results are displayed as the tests are running and
* appended to a log file, BusRead.txt. The benchmark runs via Win9X or
* NT/2K/XP. "

For this test, I didn't see much of a difference in performance between having ReadyBoost and not having ReadyBoost beyond a natural random variance. This is expected since the data the program is testing shouldn't be sent to such slow cache (i.e. the ReadyBoost cache).

From these three tests, it appears as though ReadyBoost slows my system for the tests I've run. This is probably due to my large system memory (2GB). How about that.

No comments: