Starting our story about HighPoint RocketRAID133 Controller, I have to admit that I am lightly late with this article now. At first, I was waiting for the hard disk drives with the storage capacity exceeding 137GB (, i.e. Maxtor D540X-4G), however, very soon I realized that I might fail to get them ever. Then, there was the whole bunch of Western Digital hard disk drive with 8MB cache and SCSI drives from IBM coming. Since I have only two hands to work with and the number of test platforms is not so high as I wish, I had to postpone my experiments on ATA/133 RAID.

After that I pulled myself together and started attacking ATA/133 RAID. So, here are the first results. And the first ATA/133 RAID controller that we tested in our laboratory appeared HighPoint RocketRAID133. Please meet!

Closer Look

At first, HighPoint used to sell its ATA chips only to third companies, which produced controller cards and integrated these chips onto mainboards (among HighPoint's customers are such well-known companies as Adaptec, Iwill, EPoX, ABIT). Now the company decided to try its luck as a controller card maker as well. The result of this policy change is the whole bunch of solutions forming a new Rocket family.

We have all these controllers at hour disposal and they take active part in our tests. I am going to review and cover in great detail each of them (for instance, the review of RocketRAID 404 is in progress right now). And to begin with I suggest taking a very close look at RocketRAID133. Here are the controller's official specs:

Host side interface	32bit 33/66 MHz PCI
Device side interface	ATA133 (IDE)
RAID controller IC	HighPoint HPT372A UDMA ATA133 RAID Controller
Number of IDE channels	2 Channels
Maximum number of drives	4 hard disk drives
Supported Hard drives	Up to ATA133
Supported RAID Levels	RAID 0 (2-4 disks), RAID 1 (2 disks), 0/1 (4 disks), JBOD (2-4 disks)
Supported OSs	Windows 98/ME, Windows NT4.0, Win2000, Linux (SuSE, Red Hat, Caldera, Turbo)
RAID Management Tool	RAID Configuration and Management
GUI Function	RAID Configuration and Management (compatible with BIOS)
Additional Features	Bootable disk or disk array support Disk hot spare for automatic mirror rebuilding Hot-swapping failed hard drives for RAID 1 or 0/1 Automatic e-mail notification when error occurs Large LBA support capacity larger than 137GB

In our tests we used a controller with v.2.1 BIOS and v.2.3 drivers.

Testbed and Methods

Our testbed was configured as follows:

• SuperMicro 370DLE mainboard;
• Intel Coppermine 600MHz;
• 2x128MB PC133 Registered SDRAM with ECC by Micron;
• Quantum FB EL 10GB HDD;
• Matrox Millennium 4MB graphics card;
• Windows 2000 Pro SP2.

To build arrays we used 4 Maxtor D740X-6L (6L020J1) hard disk drives.

When we created arrays, stripe-block size was equal 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.

Here are the benchmarks used:

• WinBench 99 1.2
• IOMeter 1999.10.20

To test the controller performance in different RAID arrays with the help of Intel IOMeter tests we used the following patterns:

*- Recommended by Intel;
** - Described by StorageReview.com
*** - Described by Intel and StorageReview.com

Also we involved a few new patterns introduced by StorageReview.com:

These patterns should allow measuring the performance of the disk subsystem under workload typical of file- and web-servers.

Also we tested the performance of the HighPoint RocketRAID133 controller in RAID arrays of different types when the share of reads and writes changed from 100/0 to 0/100 with -10/+10 step.

And at last we tested if the controller was capable of working with Sequential reads and writes of different size in various RAID arrays.

Performance

WinBench99 1.2 for Win2000 (FAT32)

WinBench99 tests showed us ho the HighPoint RocketRAID133 owner could benefit from his controller in regular Windows applications.

To make the comparison of results obtained in RAID arrays of different types simpler, we composed the following diagrams with the results for High-End Disk WinMark and Business Disk WinMark:

As we see, the performance in Business Disk WinMark in case of RAID 0 array doesn't depend on the number of HDDs in the array, while in High-End test this dependence is evident. The more HDDs we add to the array, the higher grow the results. Although this rule doesn't work for a 4-HDD array for some reason. :)

On the next diagram, we compare the performance of a single hard drive with that of RAID 0 - RAID 1 arrays built on one or two channels of RocketRAID133.

It is evident that for RAID 0 and RAID 1 arrays the top speed can be achieved when the HDD building these arrays are connected to different controller channels.

The remarkable thing is that the performance of RAID01 array appeared lower than that of a single hard disk drive :(

Now let's see of the tests in another file system will change this tendency.

WinBench99 1.2 for Win2000 (NTFS)

Strange as it might seem, but the best results in Business Disk WinMark belong to a single hard disk drive. As for the High-End tests, the results grow up as the number of HDDs in the array increases, just like in FAT32 (see above).

Just like the previous time, RAID0 and RAID 1 arrays appeared faster when the hard drives building them were connected to different controller channels. But this time, RAID01 array outperformed a single hard drive.

Intel IOMeter

IOMeter results shown by HighPoint RocketRAID133 controller were collected in a single table for your convenience. In the table you can see only Total I/Os (that is the number of processed requests) under different workloads (queue depth):

Hm, it will be not that easy to analyze the obtained results:

1. The growing number of HDDs in RAID0 array speeds up the whole system, but this performance growth is not that high and doesn't go linearly. For example, the maximum performance of RAID0 array working under not very heavy load can be observed in case of 2-drive configuration. While in case of the maximum workload, the fastest appeared a 3-drive RAID0 configuration.
2. RAID0 and RAID1 arrays composed of HDDs connected to one channel are slower than those composed of drives connected to different channels.
3. RAID01 array appeared faster than a single hard disk drive only in Database pattern.

On the whole the situation in these patterns repeated. Neither in RAID0 nor in RAID01 we could detect any tangible performance improvement.

Now let's see, how the controller will work in a pattern where the ratio between writes and reads changes:

At first we suggest turning to RAID0 array and its performance depending on the number of writes under three types of workload (1, 16 and 256 requests):

As we can see, the advantages of RAID0 array can be seen only when the share of writes exceeds 30%. And then, the higher gets the number of writes coming to the controller, the more evident become the advantages of the arrays built of more HDDs.

As the queue depth increases, we can watch a very interesting phenomenon: in RandomRead mode (100% reads / 0% writes) the RAID0 array speed drops as the number of HDDs building it brows up. The highest performance was shown by a single HDD, and the lowest - by RAID 0 of 4 HDDs.

However, when the share of writes starts growing, the array speeds become more or less familiar to us (the more HDDs form the arrays, the faster it is).

Increasing the queue depth up to 256 requests moves the spot where RAID0 array defeats a single hard disk drive 10 point to the right. Other than that, the situation repeats the previous two graphs.

And now let's consider the performance of RAID0 and RAID1 arrays built on one and on two controller channels:

No doubt, the arrays built using 2 channels work faster than those built using 1 single channel of the HighPoint RocketRAID133 controller. Though I got very much interested in the RAID1 array built using 2 channels, which appeared slower than the same array built using a single controller channel in two cases! But this is a question to HighPoint then :)

As the queue increases up to 16 requests, the oddities about RAID1 2Ch array disappear. Here we have every reason to state that arrays built using 2 controller channels are indisputably faster than those built using 1 controller channel.

When we get to 256 requests queue, one more interesting phenomenon crops up. Please, note that when the writes exceeded 50%, RAID0 1Ch array outperformed the same array built using 2 channels!

RAID1 1Ch also yields to its 2-channel counterpart in case of small writes, but as this number increases, it nearly catches up with the dual-channel RAID1 array.

Now it's high time we compared how well the arrays can work with sequential reads and writes. The table below contains the transfer rates from the arrays for data packs of different sizes.

To make the picture more illustrative, we made the following diagrams:

As we see, the read rates from RAID0 array made of different number of HDDs are not that greatly different. The read rate from the single hard drive makes 39MB/sec, but from two drives building a RAID0 array it makes only 55MB/sec. And where is the promised doubling of the rate? And the most discouraging thing is the fact, that adding another, third, HDD doesn't have any positive effect on the performance at all! And only 4-HDD array manages to overcome the 60MB/sec bar. Well, what we see is evidently called scalability problems… But who is the one to blame: the controller or the HDDs?

If we compare the performance of RAID1 arrays and that of a single HDD, we will see that RocketRAID133 controller doesn't show any performance drop during mirroring.

RAID01 array appears a pretty curious sight. When the requested data pack is relatively small, it proves faster than a single HDD and both RAID 1 arrays. However, when the data pack reaches 4KB-8KB, it suddenly cedes to the rivals. As the data pack gets bigger after that, RAID01 again takes the lead and retains it until the very end.

If we compare the performance of RAID0 arrays built with 1 and 2 channels, we will see that the read rates from the array are almost similar in both cases. However, RAID0 2Ch is slightly faster when working with 4KB and 128KB+ blocks of data.

Strange as it might seem, but HighPoint RocketRAID133 controller managed to cope with writes much better than with reads. Although I suspect that the reason for that hides somewhere deep inside the firmware of Maxtor 6L020J1 hard disk drives…

It is very interesting that all RAID0 arrays performed equally fast here independent of the number of HDDs, unlike SequentialRead pattern. In case of 2 HDDs already, the write rate makes 80MB/sec and hardly gets any higher further on. Also take note that the maximum performance in all arrays is observed in case the data pack we are writing into the array appears of the same size as the stripe-block.

Hm, this is a very interesting picture. You can clearly see, that the controller workload increases when you need to duplicate the data onto both drives of the array, so the results shown by RAID1 array working with small data packs appear lower than those of a single hard drive. However, as soon as the data pack reaches 64KB, the array performance grows up to its maximum. The graph for RAID01 array working with small data packs repeats RAID 1 1Ch graph, but in case of 64KB data pack the performance suddenly jumps up. Although, further on, as the data packs get larger, the performance of RAID01 drops even below that of a single HDD.

When we compare the transfer rates of RAID0 arrays built with 1 and 2 controller channels, we see that they differ only in case the data packs transferred are smaller than 64KB.

Conclusion

Well, the first tests of ATA/133 RAID controllers were full of surprises. HighPoint controller showed excellent performance in WinBench (Business Disk WinMark and High-End Disk WinMark), however, the linear read graphs will turn out a disappointment for those of you who are fond of video editing and the like. Maybe these are the hard disk drives (Maxtor D740X-6L) that are to blame for that… But can you suggest any other HDDs to use with an ATA/133 RAID controller?

I decided to double check that is why I will also test HighPoint RocketRAID133 with IBM DTLA HDDs to see, if it really the hard disk drive that is guilty (as you may have already noticed, I usually use these drives to test IDE RAID controllers).

As for the Intel IOMeter results, we have clearly seen that the performance starts growing noticeably only when the share of writes is pretty high. In patterns emulating server workload, the performance gain is nearly negligible by any type of RAID arrays.

The weak point of dual-channel IDE RAID controllers appeared work with arrays composed of HDDs connected to a single channel, which we saw many times in this review.

And returning to our today's hero, HighPoint RocketRAID133, I have to stress that these results can be considered preliminary, taking into account that the company has recently released new drivers version 2.31. So, stay tuned for new tests of more HighPoint controller cards!