Guest Eugene

StorageReview's Office DriveMark 2006

Recommended Posts

...would add a lot of credit to SF in the corporate sector.

212206[/snapback]

Sorry, I meant SR, not SF. Lost the habit.

Share this post


Link to post
Share on other sites
Guest Eugene
What's the hardware configuration of Testbed 4?  I'm sure it's been mentioned before, but I haven't been around very much lately.

212203[/snapback]

The basic configuration is a pair of 3.0 Noconas, 7525 chipset, 2 GB ECC DDR, SI3124-2 SATA-2 controller, LSI-22320 U320 controller, LSI 1064 SAS controller.

Share this post


Link to post
Share on other sites

Well, my appetite has been whetted. I eagerly await the methodology main course and the data and analysis behind it. :)

Share this post


Link to post
Share on other sites
Heck standard DVD movies drag most current CPU's down to almost 100% utilization; throw in HD content, they cough when trying to run multiple applications.

You must be joking. 1GHz is plenty to software decode a DVD. Though HD is a much bigger resource hog.

Edited by Pradeep

Share this post


Link to post
Share on other sites
Heck standard DVD movies drag most current CPU's down to almost 100% utilization; throw in HD content, they cough when trying to run multiple applications.

You must be joking. 1GHz is plenty to software decode a DVD. Though HD is a much bigger resource hog.

212284[/snapback]

Indeed. I thought even a Pentium 2 could decode DVDs in software.

Share this post


Link to post
Share on other sites
Heck standard DVD movies drag most current CPU's down to almost 100% utilization; throw in HD content, they cough when trying to run multiple applications.

You must be joking. 1GHz is plenty to software decode a DVD. Though HD is a much bigger resource hog.

Indeed. I thought even a Pentium 2 could decode DVDs in software.

212320[/snapback]

My old Dual Celeron 300A did that with some hardware assistance (Rage128), if barely. IIRC pure software decoding (but hey, who's still got such a crappy old graphics card these days?) needs something like a PIII-800 to run smoothly.

Share this post


Link to post
Share on other sites

It always worked for me when I had PII's. I think the problem is more the players then the machine or dvd itself. The older not as flashy software dvd players used less system resources for flash/bang and more for the actual decoding.

Share this post


Link to post
Share on other sites
My old Dual Celeron 300A did that with some hardware assistance (Rage128), if barely. IIRC pure software decoding (but hey, who's still got such a crappy old graphics card these days?) needs something like a PIII-800 to run smoothly.

212325[/snapback]

A Celeron is no P2.

Share this post


Link to post
Share on other sites
Maybe instead of complaining I should lend a hand... want some help devising a good modern relational database benchmark, say Oracle-based, a couple of repeatable, highly parallelized workloads?  Let me know.

212205[/snapback]

Sure, we'll listen. The best way (at least for Windows-based databases) would be to get IPEAK SPT's WinTrace32 to capture a live database system... not an easy thing to do. We had a couple leads here and there to get a trace, but no go unfortunately. For multi-user applications, IOMeter (very synthetic compared to the single-user tests, which are actual replays of real usage) remains TB4's starting tool.

One of TB4's main aims is to streamline the time it takes to run through a test- a single drive can be formally tested in about 2 hours vs TB3's 10 horus. This, in turn, will allow us to examine array performance more rather than single drives. TB4 is a dual-xeon system aimed at examining performance beyond the single drive.

212209[/snapback]

Eugene,

Rather than measuring drive stats, would it be acceptable to measure time to complete a given database benchmark? The Transaction Processing Council has an industry accepted set of benchmarks for almost every database scenario you could want, along with guidelines on the proper administration of the tests, scripts and programs to generate large datasets with sufficient randomness, and of course database queries. They have several classes of tests:

1) Transaction processing

2) Web Server

3) Decision Support/Datawarehouse

Within each class, they have evolved the standard over the years, following an alphabetical nomenclature. TPC-A, TPC-B, etc. TCP-D happened to be the first Decision Support/Datawarehousing mark, and after that they just started getting cute and having the letters be a mnemonic for what it was they were testing. So TPC-W is for Web servers, and TPC-H is for Ad-Hoc queries (the A being previously used), and finally TPC-APP for application servers. Incidentally, the TPC-D Decision Support benchmark and the TPC-H Ad-Hoc benchmark are the ones that REALLY stress out disk performance, especially sequential Read/Write. I have executed several large TPC-D benchmarks in my career, and they take a fair amount of time to do well, because you usually have to tune the entire machine's hardware to get the best results. The good news is that SR's system is fixed, and a single CPU node - so we can run at the smallest scales of data.

The problem becomes in selecting an appropriate benchmark from those available. Should it be ad-hoc queries? Web hosting? Application Server? The problem is that the data layouts and system test are different for all of these, so it would be hard to build a common database and then mix the queries.

However, you COULD easily customize the Ad-Hoc benchmark, as it is a mix of queries, some using indexes, some using table scans. We can change those ratios as we see fit to stress various performance metrics of the underlying hardware. As we can guess, queries with large results sets will use table scans, and place a premium on STR. Queries that use the indexes primarily will stress I/O per second. We would have to examine the benchmark to determine how queries are (or are not) influenced by writes versus reads. The old TPC-D's really had no write testing at all, other than the creation of temp files (which we placed on RAID 0, while everything else ran on RAID 5 arrays for STR).

Anyway, take a look - it is a fair investment of time, but they are good benchmarks, and they are industry standards.

Future Shock

Share this post


Link to post
Share on other sites
Instantaneous just feels good ;).

What would you recommend for a Office Test Suite?  I think Eugene's choice of the Veritest Suite is excellent.  Firstly, it is a standard performance metric.  Secondly, it is actually representative of many common tasks.

As for its significance in the grand scheme of things, I don't think anyone on this board really believes that anything more than a modern 7200 RPM drive is going to make a difference for most people's experience, or makes recommendations based on such beliefs.  On the other hand, SR provides a tremendous service to those who are curious about observing the quantitative performance of disk drives.  People come to this site to satisfy their curiousity.  The measurements matter to us, but I don't think anyone's trying to pretend that hard drive performance is going to change the way most of us work or play.

212006[/snapback]

No. I come here to see what the fastest drive is on the market. All drives of the same size and generation are pretty much the same price. I come here to find out wich one is the fastest then I buy it.

Share this post


Link to post
Share on other sites
Guest Eugene

The graph and % change graph originally presented in the initial post have been updated. Basically, the Deskstar 7K400 was run with legacy ATA tagged command queuing enabled. As the WD Raptor has shown us over the course of the past two years, TCQ has a detrimental effect on single-user performance. This handicap placed the 7K400 relatively low in both these Office and the related High-End DriveMarks, much to the chagrin and question of many. With TCQ disabled, however, the Deskstar 7K400 performs as most expected- a very competitive drive.

For those too lazy to go back to the first page, here are the two graphs in question, revised for the Deskstar. Note that since there are no TCQ-enabled results from testbed3, there is by definition no % change result for the Deskstar with TCQ.

drivemark_office.png

change_office_2002-2006.png

Share this post


Link to post
Share on other sites
My old Dual Celeron 300A did that with some hardware assistance (Rage128), if barely. IIRC pure software decoding (but hey, who's still got such a crappy old graphics card these days?) needs something like a PIII-800 to run smoothly.

212325[/snapback]

A Celeron is no P2.

212332[/snapback]

http://www.anandtech.com/showdoc.html?i=277

You might want to reconsider that statement. The name Celeron is only a marketing tool after all.

While it is true that a rectangle is not always a square and a Celeron is not always a PII, it is true that the Celeron 300a is a PII core with less cache.

Share this post


Link to post
Share on other sites

So the 74GB Raptor with no command queueing was 1.5% slower than the average? :huh: I must be screwing something up...

Eugene, is there an archive for Testbed 3 or would it be possible to get some screenshots of past results? There have been a few times I have wanted to compare some drives not on the new testbed and it is hard to find that answer elsewhere. SR is such a definitive site on drive reviews, most sites let their storage section languish and it is hard to find good results elsewhere. For example, it would be nice to be able to give comparison advice to people upgrading from their $20-after-MIR WD drive. ;)

One comparison I really do miss being able to do, is the review on the venerable 15k.3. Not as fast as the newer 15k drives, but on Testbed 3 it held a substantial noise/heat advantage over the rest of the lineup. The 15k.4 was not a worthy successor to that drive.

Incredible work as always! The straight-forwardness of SR is refreshing.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now