A Look At SCSI-Performance: Fujitsu MAJ3364MC U160-SCSI — THG.RU

Not Hot, Not Loud — Still Out?

The question ‘IDE or SCSI’ is partly answered by costs and partly also by religion. IDE is considered to be fast and cheap, but rather inflexible. On the other hand, SCSI is very well reputed as the state-of-the-art storage subsystem, offering high performance, extensibility, best connectivity and flexibility, but all that comes at pretty high costs.

We will not try to change anybody’s mind or try and persuade to either of the two technologies. We rather start evaluating the current performance of a high-end Ultra160-SCSI on the one side and a UltraATA/100 drive on the other side. To point up the advances in hard disk technology, we included the two-year old IBM UltraStar 9ZX, which once used to be one of the very fastest SCSI drives. Being one of the first drives running at 10,000 rpm, it was load and hot. Compared to that, the new Fujitsu drive seems to be an example of sophistication.

SCSI vs. IDE: Performance

U160-SCSI is capable of transferring much more data than UltraATA/100 (160 vs. 100 MB/s). However, the disk drive plays a major role when applying those theoretical numbers to practice. The interface might be able to move impressive amounts of data, but as long as the drive(s) cannot make use of it, it will not make a difference if you use a slower standard.

Even though SCSI drives are usually not able to outperform IDE models at equal conditions (rotation speed, data density, cache memory), the latest interface standard Ultra160-SCSI offers some amazing 160 MB/s. The most important reason for having such large margins is of course the very fact that many SCSI systems run several drives at once. Thanks to intelligent controller chips and dedicated cache memory, SCSI is able to simultaneously serve several drives (no matter which type), which share the SCSI bandwidth. In contrast, IDE is using the system processor for most actions. That usually slows down Windows considerably when the disk subsystem is under intense load.

Users that don’t run software that requires high loads of and the disk subsystem, such as data base or server application or video editing software, will likely not feel any difference between IDE and SCSI. Systems used for standard Windows applications, games and multimedia stuff will do just fine with a fast IDE drive. However, as soon as you want to work with several drives simultaneously, it’s worth to start thinking about a SCSI solution.

SCSI vs. IDE: Features

For more information about this topic, please read SCSI Lives! Quantum Atlas 10K II — Ultra160-SCSI.

SCSI vs. IDE: RAID

RAID (Redundant Array Of Independent Disk Drives) used to be a domain of SCSI, but this has changed significantly lately. The disadvantage of SCSI is the clearly higher price for drives, cables and controllers. Some years ago, several companies (Promise, AMI and others) started to develop RAID controller chips for IDE drives. Today there are several products on the market, giving you the option to create fast, cheap and safe drive arrays with IDE hard disks. Due to the architecture of IDE (2 ports for 2 drives each), most IDE controllers are limited to a maximum of 4 drives (2 channels). Promise is offering a RAID-5 controller which is equipped with six IDE channels, all other solutions are limited to RAID levels 0, 1 and 0+1 (striping, mirroring and a combination of both). The advances in the area of IDE RAID are only interesting for workstations and smaller servers however. Large arrays can still only be realized and properly managed using SCSI hardware (Adaptec, DPT, Mylex and Tekram are some of the manufacturers).

In March 2000, Tom took a close look at the Promise FastTrak 66 IDE RAID controller and also made several benchmark runs. If you are interested in RAID configurations, I recommend the article Fast and Inexpensive — Promise’s FastTrak66 IDE RAID-Controller.

Latest High-Tech: Fujitsu MAJ3364MC

Fujitsu’s latest 36.4 GB enterprise drive runs at 10,000 rpm, ensuring high data transfer speed and quick access times. There are several variations of this drive available. The MAJ3364MP comes with an U2W-Interface instead of U160-SCSI. MAJ3182 is the same drive but with only 18.2 GB capacity. There is also a 9.1 GB drive by the name of MAJ3091 available.

It looks as if the power connector is missing.

The MAJ3xxxMC family comes with an 80-pin connector (SCA), which includes the power lines as well. In order to use this kind of drives in standard computers, Fujitsu ships a little adapter:

This side will be connected to the Fujitsu hard drive.

Side two hosts both the 68-pin SCSI interface and the power connector.

Now let’s take a first look at the transfer rates.

Write-performance is always lower than read-performance. When writing data to the disk, the heads need to be positioned more accurately over a track to make sure neighbor tracks aren’t affected. Very often the default setting of the write-cache is off for safety. Read after write verifying is another reason. However, the drive never drops below 20 MB/s — an excellent result, which enables it for high-end streaming applications (audio or video editing). Using multi-drive setups like e.g. a RAID configuration can speed up the transfer rates even more.

Hot Oldie: IBM DGVS UltraStar 9ZX

Two years ago we used to run our standard benchmarks with this nice IBM drive. At that time, the UltraStar 9 ZX was one of the fastest drives available. As you can see on the photo, it has a height of 1.6″. The Fujitsu drive comes with four times the capacity at only 1″ height. Today, only 72 GB enterprise drives have a height of 1.6″.

The DGVS’ seek time is still excellent even today, while the data transfer rate cannot keep pace with current models. Modern hard drives come with 2 MB or more cache memory today. The DGVS used only 1 MB, which used to be considered quite a lot two years ago.

IBM’s UltraStar 9 ZX SCSI was available only with U2W-SCSI interface. At that time, the new SCSI cables with twisted wires became widely used:

That’s an 68-pin twisted SCSI cable with terminator.

The first drive that worked at 10,000 rpm was Seagate’s first Cheetah model, which was even louder and hotter than the IBM drive. Still, running the old UltraStar 9 ZX without at least a good airflow leads to temporary malfunctions and data losses. As you will see later, the Fujitsu drive does not get as hot any more, even though it works at the same rotation speed as the DGVS. It still requires cooling though.

At the right side of the drive you can see several micro-jumpers for drive configuration.

Since the DGVS was some kind of pioneer of the 10,000-rpm-technology, it was a totally flat footed drive, as it was both extremely hot and noisy and therefore hardly suited for the usage in a desktop system.

King Of IDE: IBM DTLA DeskStar 75GXP

We already reviewed the DTLA five months ago: Fastest IDE Hard Drive Ever: IBM Deskstar 75GXP

So far there is no drive available which would be able of taking the performance crown from it. However, there are some interesting alternatives:

• King Of Capacity: DiamondMax 80
• New IDE Hard Drives at 20 GB per Platter: Fujitsu MPG3409 and Western Digital WD400

The DTLA is very fast, reliable and no rake. Even though it runs at fast 7,200 rpm, the drive does not get very hot and its noise level is also more than just acceptable. Finally, the models at 30 and 45 GB are attractively priced.

That is the 40-pin IDE interface. Please note that cables used for UltraATA/66 and ATA/100 have 80 pins, even though the connector did not change for years.

IDE means Integrated Drive Electronics. However, there does not seem to be a lot of (discrete) electronics on this one. Both SCSI drives come with a bigger board.

Test Setup

Data Transfer Rates

It is amazing to see that a two-year old high-end drive performs almost like a notebook drive today in terms of data transfer. The maximum of 16.8 MB/s is less than half the performance today’s drives are able to provide. Even the IDE drive (DeskStar 75GXP) performs twice as good.

There is a clear difference between the new Fujitsu drive and the DeskStar 75GXP. As mentioned in the article summary, I’m not quite sure if the 10,000 RPM-drive MAJ3364MC could still prevail if the DTLA was running at 10,000 RPM as well.

Disk Access Time

Thanks to the high rotation speed, the Quantum Atlas 10K II and the new Fujitsu MAJ3364MC are below 9 ms — that is an excellent result. Since the UltraStar 9 ZX has a lower data density, the access time is naturally a bit higher. The 11.1 ms scored by the DTLA are also a remarkable result for a 7,200 rpm desktop drive.

In terms of access time, the technical development has not been too huge. For example data base applications will not run much faster if you exchange a two-years old hard drive with an actual one, because those applications depend much more on fast accesses than on high bandwidth.

WinBench 99 — Disk WinMarks

Ziff David’s Application-related Benchmark WinBench 99 shows a similar picture. The red lines represent the Business WinMark, while the yellow ones show the high-end results. Again the DGVS cannot impress at all any more.

Disk Operating Temperature

Heating up to 59° Celsius, IBM’s colossus drive from 1998 is a sizzling affair which should definitely be cooled. The only one out of all four drives that I would operate without active cooling is the DeskStar 75GXP IDE. Surprisingly, the new Fujitsu drive does not get too hot if you consider the high rotation speed of 10,000 rpm.

Disk Operating Noise

Thanks to the relatively slow rotation speed, the DeskStar 75GXP IDE drive is the least noisy model — both in access and idle mode. The idle noise of IBM’s UltraStar 9 ZX is far too high to use such a drive in a desktop PC. Even if you have a solid computer case, you will always hear the high frequency whirring of this first generation 10,000 rpm drive. The new Quantum Atlas 10K II and the Fujitsu MAJ3364MC have become much better in this aspect. Still the access noise is cracking and surely bothering, but the idle noise was considerably reduced.

Conclusion

It is a strange feeling to see our former high-performance reference drive scoring almost pathetic results. Going through our numbers clearly shows that the extent of technical advance and improvement in the storage sector is at least as huge as with microprocessors.

The performance of high-end drives doubles almost every two years. That’s of course not valid for the access time, but data transfer rates increase considerably with every new drive generation.

The least expensive drive of this test is no loser at all. If you are looking for a system drive with a good price/performance ratio, any high-end IDE drive similar to the IBM DTLA series will provide excellent performance which is not far away from three or four times higher priced SCSI monsters.

Yet we should not forget that most SCSI drives are optimized for multitasking applications. Simultaneous or uninterrupted data access is the big strength of SCSI drives. Usually, the CPU load during drive accesses is less as well. Unfortunately, the SCSI controller will add some extra costs to the price of the already expensive SCSI drive.