Macintosh SCSI HD Formatter Benchmarks
There must be a reason for the plethora of formatter utilities on the market. There must be some hidden benefit in using a third-party formatter. On these pages you can find a patch for Apple's HD SC Setup that enables it to recognize almost all SCSI drives, so why should someone pay hundred dollar or more when a drive can be formatted for free?
Well, ofcourse all manufacturers of formatters will market their program for a number of reasons but the most commonly heard reason is performance. Most utilities will claim that their SCSI driver (click here for info about drivers) is faster than Apple's, resulting in better performance of your harddisk. Sounds fine.
One would expect that such a critical part of your Macintosh's operating environment gets a lot of attention from the reviewing press. In practice the number of tests done is fairly limited. A test that compares all formatters does not exist at all. I found the following tests:
So, I performed my own tests. And the results are astonishing. Read on.
The drive used was a new IBM SCSI drive (OEM DFHSS4F 4.2 Gb. Ultrastar XP) connected to the external bus of a Quadra 900. All the following tests where done using System 7.5 (which contains SCSI Manager 4.3). I initialized the disk with various utilities and then did some benchmark tests.
Addendum: some people suggested that the internal bus of a Quadra 900 offers better performance. However, when I connected the drive to that bus there were no differences.
The following utilities where tested:
- HD SC Setup 3.01 (A/UX)
- HD SC Setup 7.3.5
- FWB HD Toolkit 1.8
- Silverlining 5.7
- MicroNet Utility v7.1.9j
- Lido 7.56
- APS PowerTools 4.0.7
- FormatterOne Pro 1.4.4
- Disk Manager Mac 2.42
- Mt. Everything 1.1b1
- SCSI Director 4.04
- Anubis v2.54o
- Drive7 version 4.1.8
I formatted the drive with the mentioned utilities which was a mixed blessing. Some utilities worked smooth, others took a while to get used to. (Why do some of these formatters perform a low-level format without asking - it takes almost an hour on a large drive!)
Here are the results : (If you want more detail, click here.)
Conclusion : In this setup the difference between a recent version of HD SC Setup and commercial utilities is nominal. The old HD SC Setup 3.01 (A/UX) is too slow.
The benchmarks were obtained with FWB HDT BenchTest 1.8
The sub-tests with their explanation:
Sustained Transfer Test
This test determines a drive's maximum sequential data transfer rate. Large amounts of data are transferred, so a drive's seek time and rotational latency have little effect on this test. This test simulates tasks that read or write large quantities of data, including copying files, opening large graphic files, and saving large data files. The program performs tenty 768K reads and writes and times each transaction to determine the drive's average, low and preak transfer rates. The reads and writes occur at different random starting locations on the disk. A seek to the random block is made before reading/writing to the disk.
Transactions per second
Calculates the number of sequential single block read transactions that can be performed in a second. This test is useful in finding drives that respond quickly to bursts of read requests that the Macintosh is known to perform. The write test is similar to the read test, except writes are performed. Data is read from the area to be overwritten before the sequential writes occur.
A couple of observations remain.
- The processor in the used system is a 25 MHz. 68040. The maximum obtained sustained transfer rate (3354 Kb./sec.) suggests that this is by no means a state of the art machine. It is possible that at higher SCSI bus and processor speeds the differences between the tested utilities become larger. However, considering the fairly limited functionality of a disk driver, it is doubtful whether the disk driver ever will be the bottle-neck. Most likely the SCSI bus and controller will dictate the actual performance of the system.
One reader confirmed that on his PowerMac 8500 (fast internal SCSI bus) FWB HD Toolkit 1.8 is some 50% faster compared to Silverlining 5.7, suggesting that indeed at higher speeds differences become more noticeable.
Click here for benchmarks done on a totally different setup that are in line with my findings.
- It is possible that some of the tested utilities install both an 680x0 as well as a native PowerPC version of the driver. Since the system used is 680x0 based, any native PowerPC driver present has not been tested.
- I used a minimal System 7.5 both with and without the SCSI Manager 4.3 extension (in System 7.5.3 and higher, SCSI Manager 4.3 forms an integral part of the System file) to determine the performance gain between SCSI Manager 4.3 and the 'old' SCSI Manager. Inside Macintosh indicates that 'SCSI Manager 4.3 performs similarly to the original SCSI Manager'. Using SCSI Manager 4.3 on the Quadra 900, HDT BenchTest indicated that sustained read decreased some 7% but sustained write performance went up 60%.
So after all it does pay to use a recent MacOS version!
- The amount of disk space occupied by the disk driver varies greatly between the various utilities. Some take only 16 Kb., others take several hundreds of Kb. Something to consider when small (eg. removable) media are used.
- I also copied 2070 files, together 127.8 Mb. of data, to the drive used. The differences between the tested utilities again were marginal, even HD SC Setup 3.01 (A/UX) showed good performance (the graph above suggests that it should have been slower). With or without SCSI Manager 4.3; the time taken was more or less the same. This is common fact; benchmark programs sometimes give results that are not in concordance with the results from normal end-user operations. The diskmix benchmark of MacBench 3.0, which benchmarks a realistic mix of disk I/O requests, indicated that the maximum difference in performance between the various disk drivers on test, inclusive the slow HD SC Setup 3.01 (A/UX), was around 8%.
- The slower sustained write performance of some utilities is explained by those utilities not enabling the write cache on the test drive. An example is Disk Manager Mac.
Click here if you want to determine whether write caching is enabled in your setup.
- Silverlining 5.6.4's sustained read performance was almost 15% higher compared with Silverlining 5.7, but that version did not enable the write cache, resulting in a 16% lower sustained write performance compared with version 5.7.
- Mt. Everything 1.1b1 is not a real formatter but an excellent utility for mounting media since it contains its own driver. I wondered whether that disk driver could be used to speed up a drive formatted with an old utility. It appears not; Mt. Everything's driver is as slow as the slowest utility on test here.
- I do not understand how the benchmark of some utilities could result in a faster or slower access time than the average 12 msec.; access time should be dictated fully by the drive. Not by the software that drives it. If someone has a clue let me know.
- Mr. John Catalano of Casa Blanca Works (makers of Drive7) wrote to me:
Please be aware that few honest formatting authors nowadays sell
their products on raw speed. So many other variables and improvements
have been made to the Mac OS that (as we have always said) most good
formatters long ago reached the state of the art.
Since then, it's been mostly fine tuning.
Further confirmation that performance differences between formatter utilities are diminishing.
- Drive7 comes with a remarkable Control Panel, called Mount Cache, that really boosts the perceived performance of your harddisk. I gave it 512 Kb. cache, enabled the disk driver's cache, the system's cache as well as read ahead and write behind (dangerous!) cacheing. The results as benchmarked by HDT BenchTest were remarkable; sustained write performance doubled and the number of read and write transactions per second tripled!
The boost in real-life performance (copying large amounts of files, opening a large file) were less impressive but still some 10%.
BTW: Drive7 is based on the same driver code as Apple's HD SC Setup 7.3.5 is.
- During testing I suddenly measured a considerable drop in performance. It took me a while to figure out what was going on. It turned out I had switched on my old CD-ROM drive. This old SCSI-1, async drive slowed down everything on the bus. So it is wise to always use only SCSI-2 devices on one bus and to use always a recent driver on all your disks.
- I tried FWB HD Toolkit 2.05 to no avail. It aborted when asked to create volumes on the test disk with an error message suggesting that it could not create system resources. Someone suggested allocating more memory to no avail. I am waiting for further clues from FWB's technical support.
- As always: your mileage may vary.
Back to Ernst J. Oud's homepage