Cheetah Versus Raptor: Benchmarks

[Pleonasm]

Using data from the Tom's Hardware Guide website, I constructed a performance comparison of these two drives:

Western Digital Raptor (WD1500AD)
Interface: SATA/150
Capacity: 150GB
Cache: 16MB
RPM: 10,000

Seagate Cheetah 15K.5 (ST3300655SS)
Interface: SAS
Capacity: 300GB
Cache: 16MB
RPM: 15,000Across a range of benchmarks, the Cheetah outperforms the Raptor by:
Random Access Time (h2benchw 3.6): 18%
Average Read Transfer Performance (h2benchw 3.6): 46%
Average Write Transfer Performance (h2benchw 3.6): 69%
Windows XP Startup Performance (PCMark05): 42%
File Writing Performance (PCMark05): 35%
Workstation I/O Benchmark Pattern (IOMeter 2003.05.10): 84%My question: Why, then, is the Cheetah listed as significantly underperforming the Raptor on the Office DriveMark 2006 and the High-End DriveMark 2006 benchmarks reported on the StorageReview website (see Drive Performance Resource Center)?

All insights are most welcome.

 

[babaghan]

Gosh, looking at these numbers, I want to install four SAS Cheetahs now to my gaming system

 

[babaghan]

Maybe you're right. They probably cost an arm and a leg too

 

[geofelt]

I bought a fujitsu 15k drive and pci controller on e-bay. They are getting cheaper in the smaller capacities as server users are upgrading to larger devices. It worked fine, but I did not think it was any different from a raptor. Why is this? I can think of a couple of reasons:
1) Benchmark numbers look impressive, but they do not necessarily translate into real improvement in every day usage. Read some of the articles on anandtech and storagereview on raid-0.
2) The raptor is designed for the single user desktop environment. When you read a record, the device microcode can read the succeeding records into the cache with a fair chance that the next record can be satisfied from the cache. Scsi devices are designed for the server environment. This means mostly short random reads, and many concurrent requests. The device microcode has no reason to read extra records because there are so many other users that the records would be flushed from the cache before they could be requested. Server devices can benefit from ordered seek queueing. That is the rearrangement of i/o requests to minimize the seek time for a large number of concurrent requests. In a single user desktop environment, multiple concurrent requests are uncommon, and implementing NSQ may actually be counterproductive. See storagereview.com for more on this.
For what it is worth, I did not find that my fujitsu 15k was noisy or hot.
What I would really like to see is a sata based 15k device for the high performance desktop.

 

[Spikke]

I don't see much of a point in comparing these drives.

 

[darkangelism]

the raptors shouldnt outperform 15k drives. The storage review didnt have SAS it was using scsi.

i have a sas controller and will get 15k sas drives eventually, but they do run hot so i will be using seperate fans for each

 

[apache_lives]

True, not to mention you'll need a $500 scsi controller (there are cheaper ones, but if you go high end, you don't want to cheap out on a failing controller). But don't worry about it, I doubt you'll need much more speed than raid 0 raptors can provide, or at least you won't be bottlenecked by your hdd(s) anymore, which means any further upgrades should actually give you a signifcant boost

and what about the interface? im sure PCI would limit a raid 0 setup

 

[darkangelism]

the raptors shouldnt outperform 15k drives. The storage review didnt have SAS it was using scsi.

i have a sas controller and will get 15k sas drives eventually, but they do run hot so i will be using seperate fans for each

 

[jt001]

and what about the interface? im sure PCI would limit a raid 0 setup

There is a 15k SAS drive out there that can sustain like 135MB/s by itself(was it the Cheetah?), and most are around 100MB/s so if you go raid 0 you pretty much need a PCI-e controller no matter what. Most any SAS controller will be PCI-e or PCI-x though.

 

[Pleonasm]

Darkangelism, I agree: a Raptor shouldn’t outperform a Cheetah – and yet that is what StorageReview claims (in contrast to the benchmark data provided by Tom’s Hardware Guide that demonstrates the opposite). The reason behind the discrepancy isn’t clear.

Geofelt, note that the behavior of cache on the Cheetah is programmable: “The drive provides a prefetch (read look-ahead) and multi-segmented cache control algorithms that in many cases can enhance system performance” (see Cheetah 15K.5 SAS Product Manual, pages 13-14).

JT001, Tom’s Hardware Guide states: “The Cheetah 15K.5 is the fastest 3.5" hard drive we've ever seen. Seagate claims up to 125 MB/s transfer rates, which our benchmark results confirm. We measured 128.6 MB/s as the maximum for sequential reads, which is a world record off the medium” (see Seagate Cheetah 15K.5).

 

[slim142]

Just bought the server raptor, will show my pcmark score compared to my barracuda 7200.10

 

[Pleonasm]

I_Love_Tacos, I agree that a cost/benefit analysis might favor the Raptor. However, that is not quite the question I was asking.

I believe everyone would agree that the Cheetah has considerably more “raw horsepower” than the Raptor. Hypothetically, I could envision special or limiting cases where the performance of the Cheetah and the Raptor are the same; but, I still fail to see how the performance of the former could be less than that of the latter. And, yet, that is the outcome shown by the StorageReview benchmarks. What might account for this finding (which is contradictory to the conclusion supported by the benchmark data from Tom’s Hardware Review)?

All input is welcome and appreciated.

 

[MrCommunistGen]

The Raptor outperforms the Cheetah in desktop performance because the Raptors are optimized for a desktop utilization pattern and the Cheetah is optimized for server usage patterns. The Cheetah will KILL a Raptor in a server bench. I had a link somewhere...

-mcg

 

[MrCommunistGen]

Well... I can't find the specific link, but I read it on the StorageReview website. Try reading the articles on each respective drive (the Cheetahs and the Raptors). Also, see this link for server vs desktop performance: Link
It basically illustrates what I was saying although it doesn't explain it.

-mcg

 

[slim142]

cheetah 15.5 is $300 and only comes with 73Gbs :S

plus the adapter.... blah

I go for the raptor, I also heard that the raptor is easier to install and since is desktop hard drive, compatibility is 100% with any board.

 

[Pleonasm]

Concerning the reported benchmark performance of the Cheetah 15K.5 by StorageReview, note the contradictory information that is provided: “When testing the previous generation of Seagate's SCSI drives (Savvio 10K.1, Cheetah 10K.7, and Cheetah 15K.5), StorageReview uncovered significant performance differences when these drives were set to different predefined cache segmentation strategies through Seagate's Seatools Enterprise utility. The Cheetah 15K.5, however, returns the same scores regardless of whether the utility's ‘Performance Mode’ setting is toggled on or off” (see Seagate Cheetah 15K.5).

So, does the Cheetah 15K.5 show “significant performance differences when these drives were set to different predefined cache segmentation strategies” (desktop versus server mode), or does the Cheetah 15K.5 return “the same scores regardless of whether the utility's ‘Performance Mode’ setting is toggled on or off”? If the latter, then one of two facts must be true: either (1) Seagate - a company known for its technical expertise - isn’t properly optimizing the operation of the firmware for either the server or desktop modes; or (2) the StorageReview benchmarks are not properly reflecting desktop versus server usage pattern differences.

The StorageReview benchmark methodology is described in this article: The 2006 Desktop DriveMarks. Note especially the graph “Relative Change from 2002 to 2006 DriveMark (Office)” that shows a dramatic decline (-24%) in performance by the Cheetah 15K.4 in desktop mode from 2002 to 2006. In other words, something about the newer version of the benchmark has had the greatest negative impact upon the Cheetah drive specifically. If there is that much change in the StorageReview benchmarks when the hardware is constant, can the benchmarks be trusted?

Additionally, the graph “StorageReview High-End DriveMark 2006” shows that a Cheetah 15K.4 in desktop mode outperforms a Raptor.

The bottom line? It seems to me that the lack of consistency across the StorageReview benchmarks may be reason to question their merit.

P.S.: Slim142, the Cheetah 15K.5 comes in a variety of capacities, not just 73GB.