Earlier this year we already reviewed several models from the IBM Ultrastar 73LZX hard drive family (see our IBM Ultrastar 73LZX (Discovery 2) Hard Disk Drives Review). They didn't prove to be the fastest, but performed much better than IBM's previous generation drives (IBM 36LZX aka. DDYS). Unfortunately, the most interesting model of all, 9GB IBM 73LZX, didn't participate in that test session, so now we are going to bridge the gap.

Testing Participants

So, the participants remained the same as in IBM Ultrastar 73LZX (Discovery 2) Hard Disk Drives Review. However, now we also have the new IBM Ultrastar 73LZX 9GB marked as IC35L009UCD210. Well, you might wonder why all this ado about a single drive? It's simple:- First, we leave no live witnesses :);- Second, the drive promised to be rather fast (the reasons will be explained further on);- Third, it's rather low cost that is why this and the previous item make it twice as appealing.

The above listed reasons are quite enough to explain why we are so interested in testing this drive.

This HDD caught our attention with one very intriguing detail in its specs. Let's read them through once again:

It's clear that all the drives of the 73LZH family feature 12GB platters, but the junior model has a capacity lower than the size of the standard platter. So, here we deal with a hard drive, which had a part of its platter "cut-off" (of course, it shouldn't be understood literally :)). To be more exact, the inner tracks with lower data density were simply removed.

Like a lizard that has cast aside its tail, a HDD with a "shortened" platter is going to be faster (at random requests). The shorter the platter is, the lower is the average seek time and that's exactly the parameter that largely determines SCSI HDDs performance.

Testbed and Methods

Our testbed was configured as follows:

• ASUS CUBX-E mainboard with 1007A BIOS;
• Intel Coppermine 600MHz CPU;
• 2 x 128MB PC133 SDRAM by Hyundai ;
• Matrox Millennium 4MB graphics card;
• Adaptec 29160N SCSI controller;
• Windows 2000 Pro / SP2.

The hard drives were tested with Adaptec ASC 29160N controller. All drives were tested in Ultra160 SCSI mode. For Adaptec controller we used 1.00 driver (W2K).

We used the following benchmarking software:

• WinBench 99 1.2;
• HDTach 2.61;
• Intel IOMeter 1999.10.20.

For Intel IOMeter and HDTach tests we used an unformatted drive. For WinBench tests the drives were formatted in FAT32 and NTFS as one logical disk with the default cluster size. The WinBench tests were run four times each; the average result was taken for further analysis. The HDDs didn't cool down between the tests.

We used new StorageReview patterns in Intel IOMeter. Besides, we will also compare the performance of the top model in the 73LZX family with that of the low-end one in our DataBase pattern.

Performance

Average Access Time

We measured the average time required for the drive to access a random sector with two benchmarks, HDTach and Winbench99.

The results tell us that the 9GB model of the Ultrastar 73LZX family has the lowest average access time value of all. And if we refer to this article, we will see that the 9GB IBM Ultrastar 73LZX has the shortest average access time value of all present-day 10K HDDs! Moreover, the last champion, Maxtor Atlas 10K III, fell 0.2ms behind the 9GB IBM drive!

Sustained Linear Read Speed

Now, let's check the linear read speed at the end and beginning of the 9GB model compared to other drives in the IBM Ultrastar 73LZX family:

As you see, linear read speed in the beginning of the drive remained unchanged, while in the end it is much higher by the 9GB model than by all other drives. This proves that the "shortening" of the platter was done at the expense of inner, less-dense tracks.

HDTach 2.61

HDTach results fit into the above described picture:

The maximum read speed in about the same by all HDDs (keeping in mind the measurement errors), while the minimum read speed is much higher by the 9GB model than by all other drives. Accordingly, the average read and write speeds have improved, too.

WinBench 99 1.2 (FAT32)

As the maximum read speed doesn't differ by all the HDDs from Ultrastar 73LZX family, and the tests are run on the empty drive, we will hardly see any great difference in the results:

That's exactly the case: all the drives showed nearly the same performance (with correction for measurement errors and the impact of storage capacity onto the speed).

WinBench 99 1.2 (NTFS)

We see similar picture in NTFS, too.

All the drives perform pretty much the same, although the results are scattered in a wider range than in FAT32.

Intel IOMeter

This time we are not going to have a "big" testing spree in Intel IOMeter as the considered effect is quite perfectly illustrated with just a few tests.

The first test consists of StorageReview patterns that check HDD performance under workloads typical for file- and web-servers (the table only lists Total I/O values):

The 9GB HDD is about 10% faster than the 18GB model under smaller workloads in both patterns. At longer request queue depths, the gap gets down to 5-6%.

The graphs definitely indicate that all the HDDs belong to the same family. They all behave about the same at request queue depth ranging from 64 to 256. It is the HDD firmware that largely determines the performance under such workloads and the firmware in the tested HDDs seems to be the same :).

Under smaller workloads the HDD performance mostly depends upon its mechanics and we see that the 9GB model is beyond competition here.

The comparison of the fastest and the slowest drives in the Ultrastar 73LZX family, namely the 9GB and 73GB models, in the Database pattern appeared very interesting:

As we see, the 9GB model has the same advantage over the 73GB model at nearly any read-to-write operations ratio. It's about 16%, according to the Excel report. When the write operations share approaches 100%, we see the gap growing up to 20% in the RandomWrite mode.

With longer request queue depth, the results are growing, but the gap between the two drives remains the same.

At the maximum request queue depth, the tendency for the graphs to diverge in case of high write operations share is most clear.

Conclusion

It was quite clear beforehand that IBM Ultrastar 73LZX with 9GB storage capacity (IC35L009UCD210) would be faster than all the other HDDs of that family. But we have never expected it to be SO much faster… :)

This drive seems to be the fastest 10,000rpm SCSI HDD for today.

One more advantage of the drive is its price. It costs about $90 now and thus offers an optimal price-to-quality ratio as a not very large solution (it can find its application in web-servers, thin clients and so on).

But there is a fly in our ointment. In a private talk, IBM's employees told us that the life cycle of this HDD is going to be very short. The drive was manufactured to ease the transition from the 36LZX HDD family to the new and larger-capacity 73LZX drives. And now, as this line is selling well, there is no need any more to produce 9GB drives. Moreover, as the production cost of these drives and their selling price exclude the possibility for IBM to get high profits. So, I suggest that you hurry up, if you really want to save some money.