by FastSite
09/20/2001 | 12:00 AM
As we have already mentioned in our reviews several times, IDE RAID controllers become more and more popular. The growing competition in this market field forces the manufacturers to provide these controllers with more new features making them look more similar to serious SCSI controllers. The present review is devoted to a new controller from Promise - FastTrak100 TX2 Pro. As you can guess from its name, there aren't any revolutionary changes made to it compared with the predecessor. It differs from the older FastTrak100 by its ability to work in PCI 66MHz slot. Theoretically, it should increase the controller throughput, which used to be limited by the bandwidth of the PCI 32bit 33MHz bus. As we remember, FastTrak controllers feature two channels for 2 HDDs each. In other words, the controller allows connecting up to 4 hard disk drives. The today's HDDs are capable of transferring up to 35-40MB/sec, so those 133MB/sec provided by the regular PCI 32bit 33MHz bus may turn out insufficient for 4 drives like that. The most logical solution to this problem is the shift to a faster PCI 32bit 66MHz bus, because it requires less changes to be made to the chip than in case the whole thing is shifted to PCI 64bit…
The second interesting thing about this product hides behind the "Pro" word. These three letters mean that the package contains two Promise SuperSwap mobile racks! Of course, these mobile racks (or chassises) support HotSwap function.
Well, now let's take a closer look at the controller.
The Promise FastTrak100 TX2 Pro controller arrived in the following box:
Inside the box we found the controller card:
And two nice-looking chassises (mobile racks):
There were also two UDMA66 cables in the box, the manual and the drivers. We decided not to take any pictures of them. Instead, we examined carefully the SuperSwap device.
First of all, we would like to mention that the housing of this device is made of metal:
It is a really smart thing, because the HDD doesn't heat almost at all in there.
Inside the rack, there is a power supply connector and an 80-pin IDE cable. Also on the photo below you can see a thermal sensor "built" into the chassis. Promise SuperTrak controller cards know to take the fan rotation speed as well as the temperature from the sensors built into these chassises and to display these values in the special array state monitoring program (which we have already mentioned in our High-End IDE RAID Controllers Roundup).
To ensure that HotSwap will run smoothly and securely for both, the controller card and the HDD, the chassis housing is provided with a special connector:
As you can see, there are well over 44 pin holes there (40 pins of an IDE cable and 4 pins of the power supply connector).
And this is a matching connector in the dock-station:
But the housing wouldn't be called a real SuperSwap, if it were not for this small board installed into the dock-station:
This is exactly the electronics that allows connecting and disconnecting the hard disk drives "on the fly".
The last thing worth mentioning here is the fan fastened to the dock-station:
And the front panel of the SuperSwap chassis is equipped with special ways for the disposable air filter.
Frankly speaking, we liked these chassises so much (we haven't yet seen any better IDE chassises) that we rushed to TSMS company, which is Promise's official distributor in Russia for more SuperSwap units :)
We have to admit that it was real pleasure for us to work with SuperSwap chassises, because it saves a great lot of time and trouble preventing us from connecting and disconnecting different cables, and as a result the operating system my be changed in 10 seconds.
Hey, why are we telling you all this stuff? Just take a look at our new testbed, and you'll see what we mean :)
Our testbed, the one you saw on the photo above, was configured as follows:
In our tests we will consider the performance shown by our today's hero, of course, Promise FastTrak100 TX2, for PCI 32bit 66MHz and 32bit 33MHz and by the good old Promise FastTrak100.
The tests were run for arrays built with IBM DTLA 307015 HDDs (read ahead and write caching functions enabled).
It was not for nothing that we selected Supermicro 370DLE mainboard for our tests: it features PCI32 slots as well as PC64/66MHz slots (also supporting PCI32 cards). At first, we tested FastTrak100 TX2 controller installed into the regular PCI slot, and then we moved it to the PCI64 slot and removed the jumper setting the slot working frequency.
When we created arrays, stripe-block size was equal to 64KB (as default). 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:
The controllers used driver version 2.0 (build 6).
For Intel IOMeter tests we used the following patterns:
Well, RAID 0 seems to be the simplest way to speed up the disk subsystem.
When you have several HDDs used in RAID 0, the array size will make [smallest HDD storage capacity] x [number of HDDs in the array]. However, the performance in applications doesn't follow the same simple rule. One of the parameters telling on the array fastness is the ratio between the cluster size and the stripe block size. Each user and each application this user is working in has his own typical way of addressing the HDD determined by the file size and their usage intensity. That is why if you get the cluster and the stripe block sizes matching one another in the most optimal way, you will be sure that your money wasn't wasted.
However, we won't discuss this exciting topic here :) Our today's review is focused on a bit different things.
Anyway, if you are interested in the dependence of the performance on the stripe-block size, we would advise you to check the article at AnandTech called IDE RAID Comparison. The author of this article, Matthew Witheiler, did a really gigantic job having tested 5 cards in Intel IOMeter with the stripe-blocks of all possible sizes.
And now, let's get started. First come the tests in WinBench99.
Well, the new FastTrak100 version, FastTrak100 TX2, showed brilliant results. Moreover, the more hard disk drives were connected to the array, the greater grew the gap between our hero and its predecessor. And as for the FastTrak100 TX2 working in PCI32 66MHz, it was simply beyond any competition!
We decided not to overload the review with the bunch of linear read graphs, because most of them look very much like one another. Here are the two most illustrative ones for your reference:
Feel the difference! :-)
Now let's turn to diagrams:
As we can see, the first HDD makes the greatest contribution to the overall result. However, this time, the additional HDDs in the array do not cause any significant performance increases. It means that the requests to the hard drives are not really parallel, which prevents their potential from being used to the full extent.
The smaller gets the cluster size, the less parallel work the HDDs (because there are 16 clusters fitting into a single stripe-block). Therefore, the results shown by multi-drive configurations appeared even smaller than those shown by a single HDD in FAT32.
As soon as the files get bigger, the arrays are requested in a more sequential way, so that the results obtained for RAID 0 array start depending on the number of the HDDs in the array.
The same tendency is valid for NTFS. Although the results there are still 1.5 times lower than in FAT32.
So, judging by WinBench99 results we can say that Promise FastTrak100 TX2 controller proves faster than its predecessor, FastTrak100 (especially in 3- and 4-HDD configurations). Also the performance boosts when we shift to PCI32 66MHz bus.
Here we would like to offer you four tables with the Total I/Ops parameter for all three controllers, i.e. the number of requests processed by the controller per second. You can easily compare the controllers because the results for each particular case are placed in one line. Also you can easily evaluate how greatly the performance depends on the growth of the workload. So, go ahead:
In fact, the results shown in all the three cases are very similar. FatsTrak100 TX2 appeared generally a little bit faster, especially when working with 66MHz PCI bus. The only pattern out of the seven, where you can see the difference between the controllers with a naked eye is SequentialWrite. Let's take a closer look at it.
Well, looks quite interesting, don't you think so? FastTrak100 reaches "the top" with 2 HDDs already and then the array speed almost doesn't depend on the number of HDDs connected. And the new FastTrak100 TX2 shows the performance dependence on the number of the drives in the array as well as on the bus bandwidth. To give you a better idea of the whole thing we composed a graph showing the dependence of the controller bandwidth on the number of HDDs in the array. To make it easier we took Total I/Ops in case of the maximum workload and then translated them into the so dear megabytes per second.
Well, there are two things, which draw our attention here. Firstly, FastTrak100 showed pretty poor results. Was it because of the driver, BIOS or chip? Who knows… Secondly, it was very interesting to watch FastTrak100 TX2 in PCI 66MHz perform in 2-HDD configurations. For some reason, it appeared slower than FastTrak100 TX2 installed into a regular PCI 33MHz slot. Unfortunately, we can't find any suitable explanation for that...
Veering a little bit away from the mysteries, we have to say that FastTrak100 TX2 performed at least as well as its predecessor in this test, and in one pattern our baby simply destroyed the opponent. In SequentialWrite pattern in 4-HDD configuration FastTrak100 TX2 in PCI 66MHz transferred up to 78MB/sec! Will that be enough for your video now? :-)
Here we will take a closer look at the performance of our controllers in RAID 1 (mirroring) and RAID 0+1 (where not the drives but stripe-groups are mirrored). Both RAID types guarantee that the data will remain safe and sound in case one hard drive fails, and RAID 0+1 sometimes manages to save the data even in case two drives fail at a time.
However, there is nothing for free in this world and you will have to pay for this extra security. Duplicating the info stored on your drives implies that you've got twice as many drives in your system to built the required array. Also far not all the controller cards can do the mirroring without suffering any performance drops. On the other hand, some controllers can even profit from the fact that there is identical data on several drives from RAID 1 array (you should remember this if you have read our High-End IDE RAID Controllers Roundup).
For a better comparison we added to these tables the results obtained for a single HDD and RAID 0 array of 2 HDDs.
As we can notice, with the new Promise controller RAID 0+1 mirroring doesn't arouse any significant performance drops unlike the older FastTrak100, which performs especially poorly in High-End benchmarks.
On the whole, the situation repeats: for FastTrak100 TX2 RAID 1 is almost "free" and the results for RAID 0+1 are not much worse than those for RAID 0 built with 2 HDDs.
Now let's find out what the picture looks like for each array type:
The controllers perform very closly to one another in Business tests, and in High-End tests FastTrak100 TX2 takes the lead.
Everything remained unchanged, however the gap between the leader and the other racers became considerably smaller. Well, nothing to do about it: smaller cluster prevents the new chip from showing its best.
Actually, here the things also hardly get any different. Only in High-End test the performance difference between FastTrak100 and FastTrak100 TX2 grew even greater.
Again the results appear a bit lower in NTFS, but the newcomer retains the leading position.
Summing up we can say that WinBench99 tests proved FastTrak100 TX2 controller to be a much better solution for RAID 1 and RAID 0+1 arrays.
Also we have to point out that when working in PCI32 66MHz, FastTrak100 TX2 appears faster than in case of the regular PCI32 33MHz. Lucky you are if you have a board with PCI 66MHz slot. :-)
Again we are offering you only the Total I/Ops rates:
Well, the situation was quite predictable: in all the patterns the controllers performed similarly. However, there are two patterns worth singling out here as well. FastTrak100 TX2 gains an indisputable victory in SequentialWrite, but loses a bit to FastTrak100 in RandomRead!
Just look how greatly differ the transfer rates shown by both controllers in SequentialWrite:
The significant difference between FastTrak100 and FastTrak100 TX2 is evident. What could have caused it? We used the same drivers for both controller cards, so we suspect that is was the new PDC20270 chip that is responsible for the performance difference.
Anyway, FastTrak100 TX2 appeared simply excellent in tests. Now we should only check its ability to perform HotSwap properly :-)
According to the specs, this controller can carry out HotSwap for RAID 1 configurations only.
In order to find out if the controller can really restore the arrays "on the fly", we built RAID 1 array with the HDDs inserted into the SuperSwap housing chassises. Having formatted the array and installed the operating system, we switched off one of the drives by turning the key and removed the chassis. After that we ran Paragon Disk Wiper for the disconnected drive on a separate system. Then we returned the drive to the array and FastCheck utility, hanging in the tray, started restoring the array. In 15 minutes everything was successfully completed!
Unfortunately, FastTrak100 TX2 controller doesn't support HotSwap for RAID 0+1 configurations, this feature is typical only of TX4 version. So, if you care not only about the speed and the security of your data, but also about the possibility to restore the failed array really quickly, then go for the FastTrak100 TX4, especially since it is just a tiny bit more expensive than FastTrak100 TX2.
However, if you are looking for a solution for your desktop system, then FastTrak100 TX2 is more than enough anyway.
Promise FastTrak100 TX2 proved brilliantly fast in all the benchmarks, having significantly improved the performance of its predecessor, Promise FastTrak100. Our hero performed especially efficiently in multi-drive configurations, i.e. in those cases when the controller bandwidth matters a real lot. Also the ability to work with PCI32 66MHz is an indisputable advantage of this product, as it allows to make use of higher bus bandwidth.
Undoubtedly, a great advantage of the controller we had at our disposal (if you remember, we had FastTrak100 TX2 Pro) were the two SuperSwap chassises (the "Pro" actually stands for).