DougLite
09-02-2005, 10:39 AM
Well, if I ever get my K8 system finished, I will begin hard drive reviews. Here is the planned test setup, I will update it as more components are added. More important than the actual hardware is the planned test methods, and I will explore some of them here.
- I am not equipped to run IPEAK traces like SR, Tweakers, and AT. I plan on supplementing their data rather than competing with it. I will mostly be exploring the amount of time that a given drive takes to complete tasks, rather than mesuring throughput.
- I plan on reviewing 250GB drives. Every maker has an entry at this capacity point, and they are the most cost effective purchases right now.
- I plan on exploring multi-tasking and its impact on storage performance. I will analyze performance when the disk is assigned:
A) Only the given review timed task
B) Functioning as the Windows system boot volume and carrying out the review task
C) Windows boot volume, background tasks (MP3 playback and possibly virus scan if I an produce repeatable results) and the assigned review task.
- I will use a 160GB Seagate SATA drive as a second spindle for isolating the intended review target in scenarios A and B above.
- I am exploring several kinds of tests to evaluate a broad spectrum.
A) For a game test, UT2004 is in. Applying the latest patch disables the CD check, making it an excellent test program for measuring game loading performance and the impact of OS caching without optical storage performance impacting response.
B) I plan on a file transfer test, both large and small, and both within the disk and to another disk (probably to/from a RAID-0 array to isolate review candidates)
C) A bootup test with a common image will most likely be included.
D) An application test, possibly Acrobat reader and other applications.
- I will probably not explore heat and noise too much, as the folks at SPCR are better equipped to objectively analyze these facets of hard drive design. I will probably add my own observations and throw out sensor data "as-is."
My overriding goal is inspired mainly by Kyle and Brent's video card reviews, where the user experience is more important that synthetic benchmark results. Much of the testing will revolve around timing loading screens and the like, which I gather is what most members are primarily interested in reducing. I will thoroughly analyze results and try to get repeatability as high as possible before publishing anything. Any thoughts, comments, or suggestions?
It would be awesome if there was a program that could measure the amount of time that elapses between the launch of a gien executable and the call of a 3D API (which would mean that the game had finished loading)
While I certainly like the SR reviews, and think they are the best thing going for measuring disk performance, they have some limitations, at least in my mind, when applied to real world use:
- The test candidate is isolated, and only runs the test suite. It does not run Windows or any background tasks, a relaity for many system builders that run SLED and can't afford more than one drive. I will explore SLED performance in addition to multi-spindle performance.
- They measure the ethereal "I/O per second," a unit of measure that is difficult to understand and translate into human terms. Sure, more IOps are better, but how much better? What do more IOps do for me? I will try and answer these questions.
- I am not equipped to run IPEAK traces like SR, Tweakers, and AT. I plan on supplementing their data rather than competing with it. I will mostly be exploring the amount of time that a given drive takes to complete tasks, rather than mesuring throughput.
- I plan on reviewing 250GB drives. Every maker has an entry at this capacity point, and they are the most cost effective purchases right now.
- I plan on exploring multi-tasking and its impact on storage performance. I will analyze performance when the disk is assigned:
A) Only the given review timed task
B) Functioning as the Windows system boot volume and carrying out the review task
C) Windows boot volume, background tasks (MP3 playback and possibly virus scan if I an produce repeatable results) and the assigned review task.
- I will use a 160GB Seagate SATA drive as a second spindle for isolating the intended review target in scenarios A and B above.
- I am exploring several kinds of tests to evaluate a broad spectrum.
A) For a game test, UT2004 is in. Applying the latest patch disables the CD check, making it an excellent test program for measuring game loading performance and the impact of OS caching without optical storage performance impacting response.
B) I plan on a file transfer test, both large and small, and both within the disk and to another disk (probably to/from a RAID-0 array to isolate review candidates)
C) A bootup test with a common image will most likely be included.
D) An application test, possibly Acrobat reader and other applications.
- I will probably not explore heat and noise too much, as the folks at SPCR are better equipped to objectively analyze these facets of hard drive design. I will probably add my own observations and throw out sensor data "as-is."
My overriding goal is inspired mainly by Kyle and Brent's video card reviews, where the user experience is more important that synthetic benchmark results. Much of the testing will revolve around timing loading screens and the like, which I gather is what most members are primarily interested in reducing. I will thoroughly analyze results and try to get repeatability as high as possible before publishing anything. Any thoughts, comments, or suggestions?
It would be awesome if there was a program that could measure the amount of time that elapses between the launch of a gien executable and the call of a 3D API (which would mean that the game had finished loading)
While I certainly like the SR reviews, and think they are the best thing going for measuring disk performance, they have some limitations, at least in my mind, when applied to real world use:
- The test candidate is isolated, and only runs the test suite. It does not run Windows or any background tasks, a relaity for many system builders that run SLED and can't afford more than one drive. I will explore SLED performance in addition to multi-spindle performance.
- They measure the ethereal "I/O per second," a unit of measure that is difficult to understand and translate into human terms. Sure, more IOps are better, but how much better? What do more IOps do for me? I will try and answer these questions.