Make Home Page
Add to Favorites
XbitLabs is using Twitter
Windows Live Alerts
Having reviewed some of the already available ATA/133 RAID controllers, we decided to compare them with one another to see what they are actually worth. Namely, we will take into consideration:
- Silicon Image Sil680 ATA/133 RAID Controller (Review)
- Promise FastTrak TX2000 Controller (Review)
- HighPoint RocketRAID133 Controller (Review)
This decision seemed pretty logical to us, because all the controllers were tested in similar conditions and following the same methodology. The only results that underwent some correction were those of the HighPoint WocketRAID133 controller. As you remember, in the end of the controller review I promised that I would retest the card as soon as the new drivers come out. So, we updated the results for this controller with the new driver 2.31. And we didn't believe our eyes! The controller appeared so surprisingly fast, that it could seem that we had replaced the controller card and not just the drivers.
Frankly speaking, the controller performance grew somewhat smaller in some modes, which is actually quite a common situation. Since the performance of HighPoint RocketRAID133 turned out much better with the new driver v.2.31, we decided to take the new updated results for our comparison.
Testbed and Methods
The testbed remained the same as usual:
- SuperMicro 370DLE mainboard;
- Intel Pentium III (Coppermine) 600MHz CPU;
- 2 x 128MB Registered PC133 ECC SDRAM by Micron;
- Quantum FB EL 10GB HDD;
- Matrox Millennium 4MB graphics card;
- Windows 2000 Pro SP2.
We used the following drivers:
- Promise TX2000: driver 2.00.0.24
- Silicon Image Sil680: driver 1.0.1.4
- HighPoint RocketRAID133: driver 2.31
To build arrays we used Maxtor D740X-6L (6L020J1) hard disk drives. The stripe block size was set to 64KB. For WinBench tests we used FAT32 and NTFS file systems to format each of the hard disk drives as one logical drive of the maximum size with the default cluster. All the tests were run 4 times and then the average results were taken for the diagrams. The HDDs didn't rest for cooling down between the tests.
The benchmarks used include:
- WinBench 99 1.2
- Intel IOMeter 1999.10.20
To evaluate the controller performance in different RAID arrays with IOMeter, the new StorageReview patterns were used. They were introduced in the third edition of the HDD testing methodology.
These patterns are intended for testing the performance of disk subsystem under workload typical of file- and web-servers.
Basing upon the analysis conducted by StorageReview of the type of the disk subsystem workload in ordinary Windows applications, our colleague, Sergey Romanov aka GReY, created a pattern for IOMeter:
We'll use the pattern to find out if the HDDs and RAID controllers are attractive or not for an ordinary Windows user.
We also checked the controller performance in different RAID arrays when the sequential requests of different size.
Having collected all the numbers in a single xls-file, I tried to find the most illustrative way to present the results. However, it turned out a much harder task to fulfill that I had initially expected.
After a series of experiments I decided on a table-like presentation, however, this is far not my final choice, so if you ever get any ideas about it, please feel free to drop us a line by e-mail. Anyway, the red colour in the tables stands for the worst result of the three controllers tested, and the blue colour - for the best one.
Performance
WinBench99 1.2
For WinBench99 we are offering you only the results of the two tests: Business Disk WinMark and High-End Disk WinMark:
WinBench99 1.2 (NTFS)
As we see, the indisputable victory here belongs to HighPoint EocketRAID133 controller. HighPoint Company has been using the "Caching" driver for a long time now, but here we can watch the driver optimization in this testing set to be brought close to perfection.
The second position in this test was won by Silicon Image baby, and Promise TX2000 surprised us with the third position.
Intel IOMeter: SequentialRead
This is a very simple test: the controller receives read requests (queue=4) and the outgoing data stream from the controller is considered. The table below contains data read rates from the arrays of different types (MB/sec) and in case of different data pack size.
As we see, the JBOD and RAID0 built of 3 HDDs appeared a success for Promise TX2000. To tell the truth, it looks really not bad at all in other modes as well. Judging by the number of red cells in the table, HighPoint controller suffered the hardest times of all, and the one from Silicon Image proved fast only in case of a 4-HDD array (see how greatly the performance of Sil680 drops down in a 3-drive RAID0).
It is interesting that Promise controller managed to win in those arrays, which were built involving only one controller channel and shared the first prize with the Silicon Image controller when both controller channels are involved (note that Promise controller works faster with small requests, and Sil680, on the contrary, with larger data bits).
Also I would like to draw your attention to the fact that Promise TX2000 performance in RAID01 array appeared very critical to the ratio between the data pack and the stripe block size. The performance boost when it comes to a 256KB data pack can be explained only by the fact that in this case the requested data pack can be just ideally split between the 4 hard disk drives of the array.
HighPoint controller cannot boast any outstanding results. Moreover, it's performance turned out a way too small in RAID1 1CH.
Intel IOMeter: SequentialWrite
The only difference between this pattern and the previous one is the fact that here we do not read from the arrays but write into them.
Promise TX2000 appeared faster in case of a single HDD, and in other modes it proved successful with smaller data packs. To be fair we should say that Promise yielded a bit to HighPoint controller in terms of write rate when working with RAID0 array built of 3 or 4 HDDs. However, this performance difference appears completely insignificant when we compare it with the huge gap between these two solutions and the Silicon Image controller. The latter managed to overcome the 100MB/sec bar and caught up with the maximum possible bandwidth of the PCI32/33MHz bus.
Again Promise TX2000 controller turned out very attractive in case of smaller data packs, and Silicon Image's solution on the contrary was more successful with bigger data bits. HighPoint controller performed on the average in this pattern, but we can point out that it didn't show any evident drops (as we saw in SequentialRead pattern).
Intel IOMeter: FileServer
To compare the controllers in Intel IOMeter we use only Total I/O parameter (the number of requests the controller can process per second).
As we see, the table looks like a leopard skin :)
None of the controllers can boast an indisputable advantage over the competitors. Silicon Image controller proved the best in all modes and under all sorts of workloads, with the only exception of RAID0 mode of 3 hard disk drives. In this mode the controller slows down to the speed of a single HDD.
Promise controller doesn't look very attractive in this test: see how much red color is in its cells :(
HighPoint controller is on the contrary very successful: it managed to win in three-drive configuration (Silicon Image was beaten in this mode).
Now let's have a look at the results of mode complex modes than RAID0:
Here we will have to dwell on each particular mode:
- Two HDDs in RAID0 connected to a one channel
HighPoint controller won under all workloads. Promise controller occupied the second position and the last one in this race appeared the solution from Silicon Image. The interesting thing about it is that all controllers considered RAID0 built on one channel to be fake and hence didn't show any performance increase compared with the speed of a single hard drive. - RAID1 2 channels
When the workload is linear (queue=1), the indisputable leadership belongs to Promise controller. However, as the speed grows up HighPoint RocketRAID133 takes the lead leaving Promise baby just a few tiny bits behind. Note that the performance of both controllers in RAID1 array is much higher than in case of a single hard disk drive. It looks as if the read requests were addressed in turns to both drives of the array.
Silicon Image controller supposed that RAID1 array was built only to grant higher reliability, and that the RAID1 performance is similar to that of a single HDD. - RAID1 1 channel
Under linear workload Promise controller proves the fastest. But then the leading position moves to… no, not to HighPoint, but to Silicon Image! - RAID01
Under not very high workloads the leadership belongs to Promise controller, and when the workload grows up, the latter gets outpaced by Silicon Image controller. HighPoint product in this case works as fast as a single hard drive, no matter how sad it is.
If we take the average results for all controllers under all sorts of workload, we will get the following picture:
Intel IOMeter: WebServer
In WebServer pattern there are only reads to be processed that is why the results of the controllers tested can be slightly different from what we saw in FileServer pattern.
If we sum up the best (blue colour) and worst (red colour) results of all three controllers, we will get a very interesting picture. The victory scoring 11:3 will belong to Promise FastTrak TX2000. The second prize scoring as 6:4 will go to Sil680 from Silicon Image. HighPoint RocketRAID133 will be the last one on this list scoring 3:12.
The situation is almost the same as in FileServer pattern: Silicon Image controller loses in RAID1-2CH mode, and HighPoint controller - in RAID1-1CH mode. Promise product performed so-so: not purely blue, but at the same time, not dangerously red.
Here are the scores for the testing participants:
- Promise FastTRAK100 TX2000: 9:2
- Silicon Image Sil680: 6:7
- HighPoint RocketRAID 133: 5:13
They say that stability is a sign of mastership.
If we sum up the controller results under all sorts of workload (taking the loads as equally probable), then the picture will look slightly different:
As we see, in JBOD and RAID0 the controllers run nearly equally fast, while in RAID1 and RAID01 Promise TX2000 proved more stable than the rivals. HighPoint controller didn't cope with the RAID01 and RAID1 built on one channel, and Sil680 has evident problems with RAID1 (though RAID01 appeared the best by it).
IntelIOMeter: WorkStation
I hope that those of you who have read that far away, will be able to draw all conclusions yourselves:
Workstation: RAID1 & RAID01
Not a hard task, eh? :)
And after summing up the results under different workloads, a comparison appears as easy as ABC:
Sil680 controller is the leader in JBOD and RAID0 modes, except RAID0 of 3 HDDs and RAID01. Promise again proved stable everywhere, and HighPoint RocketRAID133 failed only in RAID01. In all other modes it appeared a success.
Conclusion
The tests of our ATA/133 IDE RAID controllers showed that RAID arrays built of the contemporary IDE HDDs can ensure not only larger storage capacity, but also faster operation. At the same time it is evident that the limitations imposed by ATA/133 architecture (in particular, two devices connected to one cable) do not allow achieving the maximum efficiency of the multiple HDD configuration.
The leadership in our test session undoubtedly belongs to Silicon Image Sil680 ATA/133 RAID controller, which performance in typical configurations, such as JBOD, RAID0 and RAID01 is about 4% higher than that of the competitors. However, it has its weak points as well: it is the mirror mode (RAID1), where it loses about 30%.
Promise TX2000 controller didn't lose more than 5%. And in RAID1 it proved an indisputable leader.
HighPoint RocketRAID133 controller with the new BIOS and drivers can boasts a significantly improved performance having caught up with the performance leaders now. However, this success didn't allow it to become the best of the best in any typical configuration. The only exception here appeared RAID0 array of 3 drives and 2 drives connected to one cable. Also the controller failed in RAID01, which is most likely to be corrected in the upcoming BIOS versions.
Summing up we can say that Promise controller is equally good for any type of arrays. If you are building a small server with two mirroring hard drives, then HighPoint will be a better choice. However, for a RAID01 array I would advise to decide on Silicon Image. In striping arrays all testing participants appeared nearly equally fast, only Silicon Image solution got a bit ahead of the rivals.
P.S.: since all the controllers tested depend a lot on the software, the new drivers release may change the whole picture at once.
