Seems like IDE and SCSI interfaces started their feud since the day IDE interface was born. Time is passing by but there is no definite answer which one of them is better. IDE drives have gone through five different ATA specs already, but SCSI ones have been evolving, too…

Today we are going to try clarifying the issue by comparing the performance of SCSI and RAID hard disk drives. In order to make chances somewhat equal, the IDE drives will be tested in RAID arrays.

There will be two groups: the first includes hard drives and RAID arrays to be used as a disk subsystem of a workstation, the second group comprises those to be used in a server.

Testing Participants

Here are our today's rivals.

The first group is represented by four SCSI drives: Seagate Barracuda 36ES2, Seagate Cheetah 36ES, Fujitsu MAN3735 and Seagate Cheetah X15 36LP. A nice bunch, isn't it? Well, every drive really deserves to be here.

Seagate Barracuda 36ES2 HDD is the only today's SCSI drive with 7,200rmp spindle rotation speed as all the other manufacturers don't produce drives like that anymore. So it is most close to the IDE HDDs both in specs and in price. Moreover, this drive is exactly positioned for the use in high-performance workstations.

The next drive, Seagate Cheetah 36ES, is positioned by Seagate as a solution for both: workstations and low-end servers.

The Fujitsu MAN3735 was included for two reasons (see our 10,000rpm SCSI Hard Disk Drives Roundup). First, it's the fastest 10,000rpm SCSI drive of all, according to our benchmarks. Second, its capacity of 73GB nearly equals that of a RAID of two IDE Maxtor 6L040J2 HDDs. This justifies our intention to compare the performance of a single SCSI HDD with an IDE RAID in FC-Test.

The last in the list, but not the least in performance and price :) is Seagate Cheetah X15 36LP. We can refer to it in no other way but as to "Hi-End", and as such, it will have to work really hard to prove its supremacy…

The second group is represented by four IDE drives (Maxtor 6L020J1) and two IDE RAID controllers (Promise FastTRAK TX2000 and Promise FastTRAK100 TX4). As we have already tested them before, we couldn't miss the opportunity to use the benchmark results once again.

The question we will try to answer in this review is what disk subsystem suits best for a desktop and for a Low-End server.

Everyone knows that SCSI drives are faster and more reliable than single IDE HDDs, but they also cost quite a sum. If we want to find the best disk subsystem, we must compare systems of the same price or of the same capacity. By changing RAID array type and the number of drives in it, we will be able to build a disk subsystem nearly equal to the single SCSI HDD both in capacity and reliability. And then we will check the price list.

To compare our testing participants with the Fujitsu drive we used the results obtained for Promise FastTRAK100 TX2 controller tested with 2 Maxtor 6L040J2 hard drives. Although they were initially intended for another article, as we will see later on they also appeared quite to the point here.

We won't introduce you to our testing methods as they have remained the same for long and you can get more info about them in any other HDD review on our site.

Performance

WinBench99 1.2

Our first benchmark is WinBench99:

The RAID arrays built of IDE drives look much better than the single SCSI HDDs. The only exception is Cheetah X15 36LP that upheld its "hi-end" reputation and proved best of all.

But Cheetah X15 36LP turned slower than the four-HDD IDE RAID in NTFS. Well, four against one doesn't seem too fair :).

The overall impression from WinBench99 results is that the IDE RAID system looks more appealing than a single SCSI HDD.

Intel IOMeter: DataBase

Now, we will check the speed of different disk subsystems when working with SQL-like requests (8KB, 100%-random data blocks).

First, let's compare RAID0 and RAID1 on the Promise TX2000 controller with Seagate Barracuda 36ES2, Seagate Cheetah 36ES and Fujitsu MAN3735.

Well, the SCSI drives feel at home in these conditions. Even Barracuda 36ES2, notwithstanding its low (for a SCSI HDD) spindle rotation speed, looks surprisingly nice in this test. It was faster in nearly all the modes than the RAID1 array of two IDE drives that have the world's "best access time" for today.

The winner here is the Fujitsu drive, followed by Cheetah 36 ES and the dual-Maxtor RAID0.

Now let's check what's going on in the second group:

Seems like the four-HDD IDE RAID arrays can compete even with Cheetah X15 36LP! But we should keep in mind that a four-drive RAID cannot be viewed as a server solution as the mean time between failures would be eight times shorter for it than for a single SCSI drive.

So we'd better look at the RAID01, especially as it shows quite good results. It's most good in RandomRead mode where all the four drives of the array are processing read requests.

Intel IOMeter: Sequential Read & Write

Now let's watch our testing participants process sequential reads and writes.

The two-HDD RAID0 is beyond competition in the first group. The single 10,000rpm SCSI drives are over 10MB/sec faster than the IDE RAID1 arrays or Seagate Barracuda 36ES2.

We see about the same picture in the second group. Even the super-fast spindle rotation speed of Cheetah X15 36LP didn't help it to get close to the speed of the four-HDD IDE RAID.

This situation can hardly change in case of write operations, but let's check it out, though:

It's rather curious that the two-HDD RAID1 can compete with the slowest single SCSI HDD (Seagate Barracuda 36ES2) only in case of requests of a certain size.

The two-HDD IDE RAID outperformed 10,000rpm SCSI HDDs when the request size exceeded 4KB.

The things in the second group go like this: the SCSI drives are in the lead when the block size is below 32KB, but when it's getting bigger, the four-HDD IDE RAID0 dashes ahead.

To sum up the results in sequential read/write tests: IDE RAID0 arrays cope with this workload much better than single SCSI HDDs.

Intel IOMeter: FileServer

Now we are approaching the realm that's favorable to SCSI. The FileServer pattern should emulate disk subsystem workload typical of fileservers.

The picture is quite opposite to what we saw in the previous section. The 10,000rpm SCSI drives are the leaders under any workload while Seagate Barracuda 36ES2 is competing with the IDE RAIDs under smaller workloads.

The four-HDD IDE RAID arrays managed to surpass Fujitsu MAN3735 and compete with Cheetah X15 36LP, which is a great achievement for IDE RAID. But, as we have already mentioned, the result of the four-HDD RAID0 is of value only in this review as no normal human being would trust the functions of a fileserver to a computer with an non-fault-tolerant disk subsystem.

Intel IOMeter: WebServer

Now let's see whether the absence of write requests in the pattern can affect the situation:

Yes, the changes are here. For example, RAID1 array showed a definite speed boost (as we remember, the drivers of the Promise controller send read requests to the hard drives of the mirror-pair in turns) and managed to surpass Cheetah 36ES a few times. A nice thing to see is the RAID1 array outperforming Seagate Barracuda 36ES2 that has the same access time (10.5ms).

The IDE RAID arrays are on equal terms with Seagate Cheetah X15 36LP and leave far behind Fujitsu MAN3735 under high workloads.

IOMeter: WorkStation

The next pattern emulates disk subsystem workload when working with ordinary Windows applications.

The 10,000rpm SCSI drives are much faster here than IDE RAID arrays. Under smaller workloads Barracuda 36ES2 even manages to outperform the two-HDD RAID arrays.

But the four-HDD IDE RAIDs perform rather well in this pattern, too. RAID0 surpassed Cheetah X15 36LP under small workloads while the RAID01 outpaced Fujitsu MAN3735.

FC-Test

And now, the most "true-to-life" benchmark of our deck: FC-Test. This test helps to measure how fast certain file sets are created, read and copied within one logical disk and from one logical disk to the other. As we have mentioned in the beginning of the article, there are only two actual rivals in this test: Fujitsu MAN3735 and a RAID0 array of two Maxtor 6L040J2 HDDs.

Let's see how good they cope with every file set:

It's clear that the IDE RAID0 performs faster than the single SCSI HDD in nearly all the modes. The only exception is reading long files.

Check the Price-List

As for the price, here are the current prices (a little rounded) of the drives and controllers considered (all prices are given in US dollar):

*- we hope you will excuse our putting into the list the price of 19160N
controller for SCSI HDDs, but not the 29160 we used in the tests.

To make the pricing comparison of different combinations easy, we sorted them out in groups according to the capacity and, inside each group, according to the price. The prices for SCSI drives include the price of the controller. If you are lucky to have an integrated SCSI controller, then it means that you have already paid for it :).

The table shows that IDE RAID looks very appealing from the price point of view, too.

Conclusion

The results of the today's comparison are rather instructive.

First, we have learned that IDE RAID arrays can compete on equal terms with single SCSI drives in a number of applications.

Second, we saw that different disk subsystems proved better in different tasks: SCSI drives are best working under their native server workloads while IDE drives in RAID0 arrays do sequential read and write very well.

Third, we discovered a coefficient between SCSI and IDE drives: one Cheetah X15 36LP equals at least four Maxtor 6L020J1 :).

It's clear that every interface has its own advantages and shortcomings. SCSI means reliability and high performance, but also high price. Moreover, it needs a costly controller. IDE means huge capacities and much lower storage cost for 1MB than in SCSI HDDs. But IDE drives are much less reliable and to build a fault-tolerant disk subsystem you need to combine them into mirror RAID arrays (and this requires at least one more drive and a RAID-controller).

We hope the comparison we have carried out will help you to choose most appropriate disk subsystem for your needs according to the three criteria: speed, reliability, price.