RUN SILENT
STORE DEEP
   
 

Silence is golden when it puts 40GB in your lap.

   
  by  Jack Fegreus      
     
 

The universal truths governing laptop systems are (1) storage capacity is by definition insufficient; (2) laptops are seldom—if ever—properly backed up; and (3) laptops are shamelessly abused as they are shlepped from office, to home, to hotel. As a consequence of these truths, the intrinsic value of any laptop system sits squarely on the disk drive. On the other hand, the pittance paid for today’s disk drives makes the extrinsic cost of the storage subsystem one of the cheapest components of a laptop. Now with their new DK23DA line of 2.5-inch disk drives, Hitachi is addressing truths 1 and 3 and, more than likely, exacerbating truth 2.

These new Hitachi drives all sport a number of technologies to insure that data is transferred on and off the drive very rapidly and that the drive can store a significant volume of data. Nonetheless, the most intriguing aspect of the new drive line is that they are all designed with fluid dynamic bearings. While the mechanical design principles of disk spindle motors are not likely to be topics of conversation around the IT water cooler, they bear significant importance to the reliability of laptop drives. And given the harsh reality of the second universal truth about laptops, that makes fluid dynamic bearings a worthy topic of conversation.

 
       
 
OPENBENCH LABS SCENARIO
UNDER EXAMINATION
Hitachi DK23DA-40F 2.5-inche disk drive
http://hitachi.com

HOW WE TESTED
HP Omnibook 6000
http://www.hp.com

OBLdisk benchmark
OBLload  benchmark

KEY FINDINGS
During high stress tests—90 I/O requests per second processed—the drive noise was inaudible.
Sequential throughput on reads doubled over the OEM drive in our laptop.

 

Conventionally, the spindle motors in disk drives use miniature ball bearings, which are packed with grease. In contrast, tape drives and VCRs have for a long time used self-pumping fluid bearings. In essence, grooves are etched on the surface of the spindle which act as a pump when the spindle rotates. This self-pumping action causes a fluid, which takes the place of the ball bearings and grease, to consolidate and create a high-pressure zone that serves to separate the load bearing surfaces.

More importantly, such a fluid bearing creates a highly dampened system that exhibits a greater tolerance to operating and non-operating shock, in particular, the specifications for each drive in the Hitachi DK23DA line to tolerate a non-operating shock of 800G for 1ms and an operating shock of 180G at 2ms. In contrast, drives designed with ball bearings can suffer damage from a non-operating shock of just 250G. To put this into perspective, the Hitachi drive should be able to take a 1-foot drop onto a concrete floor and come back for more abuse. If it can survive that kind of fall with aplomb, a few bumps in the overhead luggage bin on takeoff are of no consequence.

 
     
 

Fluid dynamic bearings are unnervingly quiet. The drive’s specification for sound power when actively functioning is a mere 3.1 Bel. That puts this drive at the level of noise in a sound recording studio. For those used to the assurance of hearing their disk drive chattering in the background as they work, the first use of a DK23DA drive can be a bit disconcerting. When we tested the drive with our load benchmark, the only indication that we had that the disk was chugging away at processing upwards of 90 I/O requests per second was a constantly glowing disk access LED that never flickered.

What’s more, this silence is forever. Unlike ball bearing spindles that get loader and noisier with use—read that as wear and tear—fluid dynamic bearings have nothing to wear out. Nothing, that is, except the seals: That’s the key technology recently mastered by the major drive manufacturers that makes drives like the DK23DA series available for the mass market—a mass market that now includes a plethora of smart consumer devices for which silence is not simply desirable but de rigueur.

       
 

Silence may be golden, but Hitachi has a lot of other highly evolved technologies in this drive to rightfully bang the drums. The density at which data is packed onto the drive’s platters is state-of-the-art. The manufacturer’s specs call for 600K bits per inch lined up in tracks squeezed to a very tight 55K tracks per inch. Naturally, dealing with data at such an areal density calls for equally state-of-the-art GMR heads and data channels—here Hitachi applies MEEPRML technology on the signals. Overseeing all of this electronic wizardry is the drive’s own 32-bit RISC processor.

All of the DK23DA drives also sport a 2MB data cache and an Ultra DMA mode 5 standard interface, which calls for a theoretical 100MB per second transfer rate. That specification is far more interesting than it is relevant. In particular, the peak internal data rate coming off of the platters is about 35MB per second. So with only the possible exception of data blasting out of cache, a typical existing laptop computer with an Ultra DMA mode 4—66MB per second—interface should see plenty of benefit from an upgrade to a DK23DA drive.

With that premise in mind, we set out to test a DK23DA-40F in an HP Omnibook 6000. Our baseline Omnibook has just such an Ultra DMA mode 4 interface and came equipped with an OEM TravelStar drive from IBM. We ran our two disk benchmarks, OBLdisk and OBLload, on both drives with eye-popping results.

 
On sequential reads, data throughput off of the Hitachi drive was double that of the Omnibook’s OEM drive.
 

In this test, the Hitachi drive—with help from that capacious 2MB cache—literally doubled the throughput rate off of the laptop’s OEM drive. While OBLdisk clocked the OEM drive delivering data at about 12MB per second, the Hitachi drive was streaming at close to 25MB per second.

Our second benchmark, OBLload, stresses disk access far more than it does data throughput. This benchmark simulates database access in a high-volume transaction processing environment. The goal is to measure the total number of I/Os that can be completed with the constraint that the average of the response times never exceeds 100ms. Alternatively, the data from this benchmark can also be viewed from the perspective of how many I/O requests per second can a system process constrained by response time.

In these tests, we saw performance improvements on the order of 25% with the DK23DA-40F. With the Hitachi drive, we were able to process upwards of 90 I/Os per second as opposed to 70 with the OEM drive. Today, this is not all that important for the majority of laptop users. Nonetheless, as journaling file systems come into broader use, this aspect of performance will take on more relevance as users experience background checkpointing.

In I/O loading stress tests, the Hitachi drive was able to service more processes making requests subject to the constraint that the average of all request response times remained under 100ms. Mouse over the chart to view the I/Os per second that could be completed versus response time.