Review: Blast off with Cenatek's Rocket Drive

Storage
Viewing page 2 of 2 pages. Previous 1 2

MORE IOMETER RESULTS

What happens with the transfer block set to 2 KB and 512 bytes is very interesting and illustrative. Shown below are screenshots from IOMeter for results for tests on both the IBM drive and the Rocket Drive with the transfer block set to 2 KB.

In case there is any question of the differences, here is a quick comparison.

IOMeter (2 KB block)
IBM HDD
Rocket Drive
I/O per second
106.76
13211.48
MB/sec
0.21
25.8
Avg I/O access (ms)
9.4
0.074
Max I/O access (ms)
51.9
14.1
% CPU Utilization
2.9%
71%

The numbers are so one-sided it's almost funny, but two things stand out for me:

  • I/O: With this block size, Rocket Drive scales up to handle over 13,000 per second, 50 times more than when the block size is 512 kb. Compare this to the poor IBM drive, which only manages 107 I/Os per second, less than 3 times higher than with the big block size. Physical drive head speed limitations obviously come into play here.
  • % CPU Utilization: Why 71%? Because the data stream is fast enough to keep the CPU that busy. From a work point of view, this is a good thing, processor cycles are not being wasted.

Now with the transfer block set to 512 bytes:

The IBM's ability to handle the smaller size block is worse due to mechanical latency, while Rocket Drive continues to ramp up. Every Rocket Drive benchmark has improved, and it is now moving the data fast enough to keep the CPU busy almost all the time.

IOMeter (512 bytes block)
IBM HDD
Rocket Drive
I/O per second
106.04
17419.41
MB/sec
0.05
8.51
Avg I/O access (ms)
9.4
0.056
Max I/O access (ms)
50.5
9.77
% CPU Utilization
2.23
94.63

OTHER BENCHMARKS + PHOTOSHOP

After simply playing around with the Rocket Drive in the XP1600+ test system for a couple days, I decided to see if any of the GENERAL benchmarking tools for desktops would show the speed improvement I experienced. This was an exercise in futility.

Many hours were wasted configuring and reconfiguring the XP1600+ system: without Rocket Drive, with Windows virtual memory in Rocket Drive, with the benchmark program loaded onto the HD, with the benchmark program loaded to Rocket Drive, etc. The benchmarks tried include: Sisoftware Sandra 2002, PCMark 2001 by MadOnion, and Content Creation 2003 and Business Winstone 2002 by Ziff-Davis / PC Magazine. None showed any significant differences in the scores. Only Sisoft Sandra was run on the P4-2.8 GHz system. Predictably, most benchmarks showed it to be almost twice as powerful as the AMD 1600+ system.

SiSoftware Sandra CPU Math Benchmarks
Dhrystone ALU
Whetstone FPU
AMD XP1600+ system
3861 MIPS
1916 MFLOPS
Intel P4-2.8 system
5557 MIPS
3501 MFLOPS

In the end I resorted to timing image manipulation operations in Adobe Photoshop 6. The original file was an 11 MB photo from a Canon G2 digital camera. The results are shown in the table below.

  • All values refer to the time required to complete the operation.
  • The hard drives were defragmented with Norton Utilities 2002 prior to these tests.
  • The aforementioned P4-2.8 GHz test system was used as a conventional high-end reference.
  • Windows virtual memory was set to be auto-managed on the hard drive in each systems.
  • All Windows settings on the two systems were set to defaults.

IBM HDD: Photoshop was installed on the IBM 75GXP hard drive of the AMD XP1600+ system and the Photoshop scratch disk set to IBM drive as well.

Rocket Drive: In the above system, the Photoshop scratch disk set to the Rocket Drive

IBM HDD: Photoshop was installed on the Seagate Barracuda IV hard drive of the P4-2.8 system.

Photoshop Operation
AMD XP1600+
P4-2.8 GHz
IBM HDD
Rocket Drive
Resize image: 11 MB to 177 MB
13 seconds
6 seconds
20 seconds*
Lighting effect on 177 MB image
50 seconds
21 seconds
35 seconds
Open 177 MB image
23 seconds
6 seconds
23 seconds
Auto Level on 177 MB image
32 seconds
10 seconds
25 seconds

The results speak for themselves. Because the scratch disk location on the hard drive is the biggest bottleneck in the conventional setup for both systems, the faster P4 shows only minor advantages over the slower XP. The Rocket Drive transforms the XP into a speed demon, with 3 of the 4 operations performed 2.5 to 3 times faster than the P4.

*It is not clear why the P4 system is slower on this operation than the hard drive based XP system. The test was repeated many times with the same result.

CONCLUSIONS

The Cenatek Rocket Drive is an unusual product that enables dramatic PC performances gains that are probably impossible to obtain any other way. Its ability to take on roles normally relegated to a hard drive but at speeds many times faster than a hard drive provides measurable and easily perceivable improvements in system performance. The IOMeter results are interesting but somewhat theoretical here because I do not have an application that can demonstrate clearly the impact of the amazing small data size flow rate of Rocket Drive compared to a hard drive, even the fastest SCSI RAID setup.

It would really shine in a database with large numbers of small files that are constantly accessed by multiple users. The latency of hard drives versus the RAM speed of Rocket Drive would mean that the drives would be churning and chugging in a desperate struggle to keep up with requests while RD would simply fly. A 10X improvement in access is not difficult to imagine for such applications: in a web server, file server, database server and numerous other industrial, commercial and research applications. This is precisely Cenatek's target market.

As demonstrated by the experiments with Photoshop, simply using the Rocket Drive for caching purposes can lead to multi-fold speed gains. Any program that requires caching, from Windows itself to web browsers and all kinds of image and video software, stands to benefit. For professional image/video production where large files are regularly worked, Rocket Drive could easily lead to real and measurable productivity gains. Anyone who has worked with Photoshop on a big poster file knows that waiting while the hard drive grinds away on a special effect or filter is virtually unavoidable -- even with multi-processor workstations that employ huge gobs of RAM. Just take the typical ~20 second gains shown in the Photoshop table above and multiply it for each Photoshop action in a large digital imaging shop... the time savings are very easy to imagine.

From the point of view of a desktop user, even a power desktop user, the pricing of the Rocket Drive is daunting. Even using the cheaper bare board plus RAM approach, for a 1G Rocket Drive, you're looking at around US$800. A 512MB will be less than $500, but that's not cheap either. If all you are wanting to do is use RAM for temporary caching, large amounts of system memory combine with Cenatek's software product, RAMdisk, will do an equally good job at a much more affordable price -- the RAM plus US$49 for the software. But if you want to load up any kind of data and permanently save it to fast RAM, RAMdisk is not the answer, because even though it can be configured to automatically load a disk image at startup and save the image to disk at shutdown, it cannot protect against data corruption or power outages, etc. Rocket Drive provides that type of security when combined with a UPS. It is a much more secure data storage platform.

For Quiet Power Computing

For quiet computing applications, the big disappointment is the Rocket Drive cannot yet replace a hard drive. You cannot boot from Rocket Drive.

In many ways, the most cost-effective desktop use of Rocket Drive is for a noise-conscious power user who has a system something like the XP1600+ test platform. It literally transformed that modest system into a monster easily competitive with and often beating the P4-2.8 system on many applications. I installed a Command and Conquer game on the Rocket Drive and was amazed at the speed with with it loaded -- about half of the usual 60 seconds needed. Game speed was little changed, bottlenecked by the slow video card. A mid-power CPU system with a fast video card, decent amount of RAM and Rocket Drive could probably be made about as quiet as my own quiet reference systems (20 dBA @ 1 meter or better) but beat out just about any high-end single CPU system in the vast majority of applications, maybe even 3D games. (Yes, it does add up to 20W of heat, but the added heat is not concentrated in a tiny CPU core and thus easily dissipated with spot cooling and low impedance case airflow.) Food for thought.

Our thanks to the good folks at Cenatek for providing us the prolonged opportunity to examine the Rocket Drive. Silent PC Review members may be interested to know that we have decided to integrate a Rocket Drive into our new web server. Yes, we were convinced!

* * * * *

Discuss this article in our Forums.



Previous 1 2

Storage - Article Index
Help support this site, buy from one of our affiliate retailers!
Search: