Click for SMOKING FAST 6G SSDs!
Initio Miles and SoftRAID Revisited - A Three Drive Array
By Mychailo Toloczko 12/9/98
If you look back to previous articles in the I/O section of Accelerate Your Mac, you'll see my previous article on setting up a dual drive array in a 7500 using the Initio Miles PCI UW SCSI card and SoftRAID. It has been many months since the dual drive array was installed, and the desire to further hotrod my machine could no longer be ignored. The dual drive array has become a three drive array.
For this article, the performance of a single drive, a dual drive array, and a three drive array were compared using MacBench 5.0. In each case, the drive or array was attached to the Initio Miles UW PCI SCSI card. Performance and reliability details follow.
Here is the setup used:
7500 (VAL-4) w/166 MHz 604e Apple card
256 MB RAM
256 KB L2 cache
one or two or three 2GB Atlas II drives (XP32275W)
Initio Miles PCI SCSI card, BIOS 1.06, PCI slot A1
IX Micro Ultimate Rez graphics card, IX 1.05 driver, PCI slot B1
5.0 MB disk cache
Speed Doubler 8.1.2 (only emulator used)
64 block (32 KB) stripe unit
Driver memory pool set for 384 concurrent I/Os
In all cases, the drives were setup as HFS+ volumes which are known to perform at about 95% of the speed of a standard HFS volume. (Sorry I don't have any links to webpages that prove this, but I believe many of the original HFS/HFS+ comparisons were posted on this site.) Mac Bench 5.0 disk scores are shown in the figure at the right. Some interesting observations are:
- Single drive: When it is taken into consideration that the test system utilized a 166 MHz 604e processor, I think its fair to say that the single UW drive CAN perform as well as the G3 IDE setup. (Personally, I've been quite suprised at how well the G3 IDE drives have performed, as IDE drives have always been touted as substandard in performance compared to a SCSI drive.) The single drive configuration also shows that the max sustained transfer rate of the Atlas II drive is a very respectable 9.5 MB/sec.
- Two drives that each have sustained transfer rates of 9.5 MB/sec do not yield a dual drive array that has a max sustained transfer rate of 19 MB/sec. Actually this is a well known phenomenon that is caused by the overhead and timing associated with making the two drives work together.
- The sequential read score is the only score that decreases (very slightly) as more drives are added to the array.
- The publishing disk score shows no improvement when adding a third drive to the array. This seems very counter intuitive when considering that publishing type work usually consists of reading and writing large files. The 1 MB read/write scores clearly show that adding a third drive to the array results in a big boost in sustained transfer rates when reading and writing large files.
A Newer Tech 333/166 MHz card arrived too late to be used in the controlled tests, but the MB5 disk mix test was run on the partially filled three drive array with the Newer Tech card installed. In switching from the 166 MHz 604e to the 333/166 MHz 750, the MB5 disk mix score jumped from 1121 to 1298, which is a 15% increase. (More power Scotty!)
The three drive array has been in operation for less than a week, so no long term reliability report can be offered. However, in the few days it has been in operation, it has worked flawlessly. It has been run with OS 8.1, OS 8.5, OS 8.5.1, a 166 MHz 604e, and a 333/166 MHz Newer Tech G3. It has been used with Dantz Retrospect Express 4.0 and 4.1a. The volume has been indexed for fast "find by content" searches using Sherlock. Virtual PC 2.1.1 has been used several times. Past experience with these programs has suggested to me they put a large strain on the I/O system, but with the three drive array described here, not a single crash or corruption of any sort has been observed.
It was mentioned in the previous article, but it is worth noting again that drives attached to the Initio card cannot be used as startup drives when VM is turned on (or RAM Doubler is installed) and the swap file is on the drive attached to the Initio card. Details are on Initio's website.
I dearly wanted to make the third drive fit inside the 7500 case, but the only place it could be fit was on the left front interior metal panel. It probably would have worked fine, but it just seemed a little too "Mickey Mouse" and crowded for my taste. Instead, an external HD case with UW hardware was purchased, and the drive was attached to the external connection on the Miles SCSI card. The case (an APS Pro Enclosure) and cable, which are both of very nice quality, were purchased from APS. The case has a large fan and digital active termination which were both features deamed as requirements. In keeping with the SCSI maxim of using the shortest possible SCSI chain, a 12" cable was purchased. Details of how to configure the drive jumpers are shown in my previous article on the dual drive array.
The drives were formatted using the "long initialization" routine in SoftRAID. After formatting, SoftRAID automatically installs the disk driver with the driver settings in the default configuration. The disk driver was reinstalled with the number of concurrent I/O's reset to 384. According to the SoftRAID manual, this can improve the performance of a three drive array when working with a large number of small files (ie, such as when performing a search or if the machine is used as a server). Creating the three drive array volume from the formatted drives is as simple as selecting the three drive icons in the SoftRAID "drive" window and simultaneously dragging the icons over to the "volume" window. A 64 block stripe unit was choosen because I do not deal exclusively with large files, and a 64 block stripe unit can boost small file performance slightly.
I welcome your comments. Additional suggestions and comments you wish to share with others through the xlr8yourmac site are best sent to Mike.
Copyright © XLR8YOURMAC.com, 1998.
Disclaimer: All brands mentioned are trademarks of their respective owners.
Users of this site must read the terms and conditions of use.