IBM was not itself in the fall of 2002. As a rule, Big Blue used to start selling its hard drives a little later than the competitors, but this year IBM was quick and efficient. Both HDD families from IBM, IDE Deskstar 180GXP and SCSI Ultrastar 146Z10, hit the market alongside with the new products from Seagate and Fujitsu. Moreover, they started selling at "right" prices from the very beginning.

We wonder what provoked this sprightliness: either IBM's HDD division decided to become profitable at last, or the company was going to enlarge the product range before selling this division to Hitachi?

At first we even thought that the Ultrastar 146Z10 drive had been developed by Hitachi and IBM was just a manufacturer… But after reading through the datasheets of Hitachi DK32EJ and IBM 146Z10 hard drives, we had to withdraw our supposition :.

Anyway, we got the specimens of the drive into our lab and sat down to tests immediately.

Closer Look

Here's IBM Ultrastar 146Z10 in the flesh:

Looks like any other hard drive from IBM, doesn't it?

   

But if you take a closer look, you may notice one difference. The upper cover of the case is not smooth as by all the previous HDD models from IBM, but shaggy. Maybe IBM engineers decided to use the "sharkskin" effect for better heat dissipation?

So, what hides beneath this "skin"?

You can see from the table that IBM Ultrastar 146Z10 drives follow the six-platter design, traditional for SCSI HDDs from IBM, but the platter density has been doubled compared with Ultrastar 73LZX. The maximum capacity of the drive in this family is 146GB (six platters, 24GB each). The second crucial difference from 73LZX is the twice as large cache-buffer: 8MB. The improvements in electronics don't end up here, however. Like all SCSI drives that tend to deserve being called "modern", IBM Ultrastar 146Z10 HDD is equipped with the most progressive Ultra320 SCSI interface. For those, who are still afraid of the number "320", IBM will be producing 146Z10 drives with the Ultra160 SCSI interface for some time.

The last remark: look at the 18GB drive in the Ultrastar 146Z10 family. It uses one "shortened" 24GB platter and, judging by our previous experience, this drive is going to be very fast!

For comparison with the new drives from IBM we will take…well, yes - the drives from IBM of the previous generation, those from the Ultrastar 73LZX family.

For more information on the previous Ultrastar 73LZX family products, please check also the following reviews:

• IBM Ultrastar 73LZX (Discovery 2) Hard Disk Drives Review
• IBM Ultrastar 73LZX 9GB Hard Disk Drive Review

Testbed and Methods

The hard drives were tested with two interfaces: Ultra160 SCSI and Ultra320 SCSI. For Ultra320 SCSI tests we used the only currently available controller, Adaptec 39320D.

For WinBench tests the arrays were formatted in FAT32 and NTFS as one logical disk with the default cluster size. The WinBench tests were run five times each; the average result was taken for further analysis. The HDDs didn't cool down between the tests.

We used the following benchmarking software:

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

To evaluate the controller performance in Intel IOMeter, we used the new StorageReview patterns. They were introduced in the third edition of their HDD testing methodology:

These patterns are intended to measure the disk subsystem performance under workload typical of file- and web-servers.

Our colleague, Sergey Romanov aka GreY, developed a pattern for Intel IOMeter basing on the StorageReveiw's study of the disk subsystem workload in ordinary Windows applications. The pattern was based on the average IPEAK statistics StorageReview provided for Office, High-End and Bootup work modes.

The pattern serves to determine the attractiveness of the HDDs for an ordinary Windows user.

Well, and in the end we checked the ability of the drives to work with sequential write and read requests of variable size.

Our testbed was configured as follows:

• ASUS P3B-F mainboard;
• Intel Pentium III (Coppermine) 600MHz CPU;
• 2 x 128MB PC100 ECC SDRAM by Hyundai;
• IBM DPTA 372050 HDD;
• Matrox Millennium 4MB graphics card;
• Windows 2000 Pro SP2.

The controllers were tested with the following drivers and BIOS versions:

• Adaptec 29160N - 4.10.4002 driver, 3.10.0 BIOS;
• Adaptec 39320D - 1.0.000.000 driver, 4.00.0 BIOS.

The reviewed drives had the following firmware versions:

• IBM Ultrastar 146Z10 73GB - S21E;
• IBM Ultrastar 146Z10 146GB - S21E;
• IBM Ultrastar 73LZX 18GB - S5BS;
• IBM Ultrastar 73LZX 36GB - S5BS.

Performance

HDTach 2.61

Let's see what we have got in HDTach:

And in fact, we have got very interesting results. First, the top-end model in the 146Z10 family has the highest access time. We have already got used to it, actually. Second, the IBM 146Z10 drives of different capacities showed different maximum read-from-buffer speeds! Both: with Ultra160 SCSI and Ultra320 SCSI interfaces. Moreover, the gap between the drives is simply huge in case of Ultra320 SCSI interface…

We can't think of what could have caused this phenomenon, as both 146Z10 HDDs have the same firmware version: S21E.

If we compare the drives of the two generations, Ultrastar 146Z10 will be the winner.

Higher per platter data density by 146Z10 helps it to outperform 73LZX in both: average read and write speeds. The maximum read-from-buffer speed by 146Z10 depends upon the SCSI protocol used and …the stars in the sky, so we won't dwell on this parameter.

Let's consider the average access time as measured by HDTach. Of course, HDTach uses fewer "random" read requests thus providing less precise average access time measurements than some gurus might wish, but we think it suits as an express test rather nicely.

And this express benchmark showed that the 73GB drive boasts better average access time than 73LZX drives, but the top-end model in the 146Z10 family appeared about 0.5ms slower than them. It looks all right considering the higher weight of the heads package (there are 12 of them in total!) in the 146GB drive. Such a package requires more time to speed up, to slow down and to position the heads to a track.

Yes, that's the price to be paid for the capacity… But the multi-platter drive has its own advantages, which we are going to reveal during the tests in Intel IOMeter.

WinBench99 1.2

WinBench tests are rather useless for SCSI HDDs, but we will still run them hoping the results will encourage you to use these drives in your home systems. :)

If we compare the results of the 73LZX 36GB HDD with the 146Z10 73GB on Ultra160/Ultra320 SCSI, we will see that the new IBM drives from the Ultrastar 146Z10 family are faster, which we expected to be the case.

Curiously enough, the 146Z10 with 146GB storage capacity was slower than its 73GB mate in the Business Disk WinMark test.

There is nothing to make lengthy comments on. The results don't greatly differ from what we have already seen above.

Have a look at the liner read graphs for the new HDDs:

• IBM Ultrastar 146Z10, 146GB (Graph);
• IBM Ultrastar 146Z10, 73GB (Graph).

If we look at the linear read graphs, we will see that the size of the first zone (the leftmost "step" in the graph) is so small, that it seems to have been made solely to "notch up" the 65MB/sec data transfer speed. :)

Well, the higher data density and larger cache buffer allowed IBM Ultrastar 146Z10 to easily leave behind the previous generation drives from IBM. Let's see if they repeat the trick in Intel IOMeter tests…

Intel IOMeter

First, let's check the performance of the hard drives in the semi-synthetic DataBase pattern. Why semi-synthetic? Because although we use 8KB data blocks (the standard page (request) size in MS SQL 7.0), we measure the speed of requests processing in a wide range, and not in some set "average" range (for example, when the read-to-write requests ratio is set to 2:1). This approach allows us to estimate the quality of lazy write algorithms in the given hard disk drive. And the five different workload values (we change the request queue depth) allow checking the effectiveness of Tag Command Queuing (the "ability" of the drives to change the order of commands processing to achieve highest overall performance).

This test produces tons of data, but we would like to offer you only the gist.

The next three diagrams show the dependence of the request processing speed on the write requests share under three different workloads. To simplify the pictures, we will show only the results of three drives tested: IBM Ultrastar 73LZX 36GB, IBM Ultrastar 146Z10 73GB and IBM Ultrastar 146Z10 146GB.

In case of request queue depth equal to 1, the 73LZX HDD demonstrates the highest speed in most modes. It is followed by IBM 146Z10 73GB. The IBM 146Z10 146GB is in the last of the three.

Under such workload (one out-coming request), the drives with lower access time perform better when the reads are prevailing. But IBM Ultrastar 146Z10 drives make up for it when the write operations share increases. The larger cache buffer makes them highly efficient in lazy writes, thus increasing the average request processing speed.

Note, though, that the IBM Ultrastar 146Z10 drives of different capacities respond to the increase in the write operations share differently.

When the request queue depth is 16 requests long, we see a different picture. The IBM Ultrastar 73LZX drive is leading in most modes. The graph for IBM 146Z10 146GB lies a bit lower and gets to the very top only in the RandomWrite mode.

The IBM 146Z10 73GB HDD loses to its bigger brother nearly everywhere. Actually, this could be explained like that: the more platters - the more chances to link the requests into the shortest chain, so that the processing of these requests could take as little time as possible. Because the more platters the drive has, the higher is the probability for it to make a "short seek", that is, the probability that the next requested data block would require the HDD to move its heads a short distance.

But this explanation doesn't suit here. Look: in the RandomWrite mode, the 146Z10 73GB HDD loses to the 73LZX, which has a twice as small cache buffer!

The moment of truth comes at the 256 requests queue. The two IBM 146Z10 drives of different capacities behave completely different, when the write operations share is on the rise! As if they were perfect strangers to each other! But they have the same firmware version!! We can see that the graph for IBM 146Z10 146GB resembles that for IBM 73LZX 36GB (if we take into account different platters, caches and so on). But the IBM 146Z10 73GB is so different from the other two that we start doubting its origin, really. :)

Now let's see how well our drives will cope with sequential reading and writing. For a better comparison, we chose the drives, which performance differed in some way from what their family-mates showed. That's why there are fewer drives in one diagram than in the other.

Here we have IBM Ultrastar 73LZX 36GB and 146Z10 with the both SCSI controllers.

As we see, in Ultra160, IBM drives of two different generations perform about the same, but when the data block size increases, the new drive takes the lead.

But the main surprise here is the sequential requests processing by 146Z10 on Ultra320: the Ultra320 HDD is the best starting from 4KB blocks! We wonder what's going to happen with writes then?

Nothing good, really. Both 146Z10 drives proved very slow with small blocks on Ultra320 controller. Note also that the drives are less speedy here writing very large data blocks, too. Curiously, IBM146Z10 HDDs of different storage capacities behave differently here once again.

What do you think, who is going to be the winner in FileServer and WebServer tests?

This time we decided to put the diagrams off and limit ourselves to the numbers only. In the "P.R." row (Performance Rating) we put the value calculated as the arithmetic mean of the Total I/O values under five different workloads.

It's clear that IBM Ultrastar 146Z10 lost to 73LZX under all possible workloads. :(

And we really can't find a good explanation to that. The data density is higher, the cache buffer is bigger, but…

Note that with the workload increase, the speed of the Ultrastar 146Z10 146GB is growing faster than of that of Ultrastar 146Z10 73GB. This must be the influence of the multi-platter design, which we have already mentioned above.

Now -comes the WorkStation pattern.

The WorkStation pattern supposes another workload range, so we calculated the Performance Rating in another way:

P.R.= Total I/O (queue=1)/1 + Total I/O (queue=2)/2 + Total I/O (queue=4)/4 + Total I/O(queue=8)/8 + Total I/O (queue=16)/16 + Total I/O (queue=32)/32

Strange as it might seem, but the new IBM drives are a little better here, than in the previous tests. Regrettably, they did lose to 73LZX, but not very much: only about 2.5%.

Conclusion

The IBM 146Z10 HDDs did rather well in our benchmarks, but, unfortunately, confirmed the emerging trend: the new drives boast higher storage capacity, but are slower than their predecessors (or sometimes they are just a tiny bit faster). The race for capacity is all right, but users would like to get the speed boost from the new products as well.

In this case, we see that the linear read speed grew up to 65MB/sec, but the performance of the drives in Intel IOMeter remained the same (or even a little worse) compared to IBM Ultrastar 73LZX drives.

As for the tested IBM Ultrastar 146Z10 HDDs, let's indulge them: they are the fledglings that can't fly yet. Let's wait until the mass shipment of the new drives and carry out our tests once again. We guess the HDD firmware will be better than in our samples. The evident problems with the Ultra320 SCSI interface and different behavior of the 146Z10 HDDs with allegedly the same firmware tells us that the firmware requires polishing. Well, we can see by the example of these very drives that the firmware is constantly being improved. The 146Z10 146GB HDD was made a month after the 146Z10 73GB and the difference is evident in DataBase and SequentialWrite patterns.

To conclude our review, let's add two points yet:

• The six-platter design of the new IBM drives means that their prime cost will be higher than that of the competitor products. So, the HDD division of the Big Blue can't hope for high profits in near future.
• IBM Ultrastar 146Z10 drives appeared in the market alongside with the new drives from other manufacturers and cost the same as the IBM drives of the previous generation. So, IBM has all the chances to keep its market share now.