AMD Contest Group
Final Round FiringSquad-Intel Edito...
Round 3 Editors Challenge Sponsored...
Top 10 Challenge Round Sponsored by...
Editors Challenge Sponsored by Inte...
FiringSquad Editors Challenge Round...
Lord Of The Rings Online Beta Conte...
FiringSquad Editors Challenge Round...
||17 entry(ies) in this category
| Your Hard Drive's Kung-fu is Weak! (37 comments )|
by: CanadaDave (303) | Posted in cluster Editors Challenge Sponsored by Intel Round 2
Posted 75 months ago ( edited 75 months ago ) in category DEFAULT
|» MEDIA (6)|
Figure 1 - IOMeter using 32KB 100% Read Setting
Fig. 2 - Galactic Civ II, New Game with defaults chosen
Fig. 3 - HOMM5, Single Player Island of One\'s Own
Fig. 4 - Company of Heroes Skirmish Map Route N13
Figure 5 - Heavy Load Comparison RAID vs. Single
Figure 6 - Heavy Load Comparison: Single Disk vs Separation
The marketing people have lied to you.
As shocking as that statement is, there is clear evidence that one of the most often-hyped features included on higher-end motherboards today may be of absolutely no value to gamers. Indeed, the marketing machines of major hardware manufacturers have been churning out factoid after factoid about RAID-0 being an “enthusiast-class” feature, the implication being that gamers can use this for an extra edge in gaming. Indeed – there are benchmarks which show very clearly that two disks can transfer data more quickly than one:
(See Figure 1)
This synthetic benchmark shows the performance difference between a two-disk RAID-0 configuration and a single disk in my test-bed system: unsurprisingly, a two disk RAID-0 array has roughly twice the throughput of a single disk. The problem with this is that the conditions found in that benchmark are diametrically opposite to what gamers today demand. Games (and their levels) are generally comprised of large amounts of small files, rather than a single contiguous file that is loaded once into memory.
Benchmarks strewn about the Net confirm the disparity between this method of testing and the reality of modern gaming. Google “RAID and gaming”, and you will be treated to a dozen benchmark articles that admonish users for wasting their hard-earned money on a RAID-0 configuration.
All is not lost, however. The concept of improving disk throughput is an important one. Rather, the application of the concept needs to be done more effectively to help gamers improve their experience.
In other words: You have not wasted your money on the second disk in your RAID-0 configuration. You've just been told to configure it incorrectly.
This article will illustrate real-world situations involving a single hard drive, two hard drives in a hardware RAID-0 configuration, and – most importantly – two single disks configured to perform separate functions.
Dell Precision 380
Intel Pentium D 3.0 ghz processor
Integrated Intel 82801 SATA RAID Controller
3 x 1gb DDR2 ECC SDRAM
NVIDIA QuadroFX 3400
2x Maxtor DiamondMax 10 80gb SATA150 hard disks (6L080M0)
Windows XP Professional x64 Edition Service Pack 1
For the tests below, the OS and drivers were installed and updated to the latest level. WinZip was installed (for heavy-load testing), followed by one of the games. For each game, a particular methodology was chosen, and timed with a stopwatch. Due to the human factor involved in controlling the stopwatch, anything smaller than 2/3 of a second in time difference should be considered to be a tie.
Each test was repeated three times in each configuration (rebooting after each test was run), to validate the results.
The system was then wiped clean, and rebuilt in the next configuration.
There were three hard drive configurations used for this benchmark:
Single Disk: A single disk was used for the operating system and game installs.
RAID-0: A hardware RAID-0 volume was set up using two disks. Both the OS and the game installs were done on the single RAID volume.
Separation: The operating system and WinZip were installed on one disk, and the game installs were done on the other.
The retail version of each game was chosen, with the latest patches installed that were available on the respective game's website.
Test One – Galactic Civilizations II. A new game was started as the humans, with defaults selected. The stopwatch was started after the last of the "Next" prompts, and was ended when the Quarterly Report screen appeared.
(See Figure 2)
You can see above that there is little to no difference running Galactic Civilizations II under any of the above configurations.
I had the chance to speak with Cari Begle, who is the lead developer for Galactic Civilizations II for Stardock Games. She confirmed that GC2, like the majority of other strategy games, load resources on demand to support gameplay. This results in a lot of drive seek time but is easier on system memory. She continued:
"We did consider putting all the individual resource files into a larger, single file, but that would have resulted in a massive memory footprint. It is far easier for strategy games to load files on the fly, as needed, rather than loading unneeded files and wasting more system resources."
Thanks to Cari for the insightful comments, Kristin for arranging the interview, and Stardock Games as a whole for their participation!
Test Two – Heroes of Might and Magic 5, Collector’s Edition. A single player game was started, using the Island of One’s Own Map with the default selections. The stopwatch was started when "Create Game" was clicked, and ended when the “press any key” message was displayed.
(See Figure 3)
Once again, there is little to no difference launching a Heroes 5 map under any of the above configurations. The increased throughput of the RAID array is unable to compensate for the disk seeking that is required to load the level.
Test Three – Company of Heroes. A skirmish game was chosen, using the map Route N13. The stopwatch began after "Start Game" was clicked, and ended when the "press any key" message appeared.
(See Figure 4)
In this game, we begin to see some slight divergence in load times, with the “separation” configuration edging the other two out slightly. Due to the massive memory footprint of this game – approximately 900 megabytes, according to Task Manager – the game required some of the resources found on the operating system disk, which was separated from the game disk under this configuration. Since the system was able to make these requests without having to seek from one area of the disk to the other in rapid succession when altering between these tasks, the level loaded more quickly.
The above three tests illustrate clearly that there is very little that a gamer can do to improve their lot in life if their only function on the PC is to boot up, load a game and play. Even the minor differences noted under Company of Heroes is hardly enough to justify the increased expense of a second hard drive.
The issue at hand, however, is that few gamers run only the single application at a time. Many of us have background applications running – their favourite BitTorrent client, IM software, virus scanner, indexing, encoding… and the list goes on. All of these additional applications produce additional activity on the hard disk they are using, which plays havoc on disk availability.
An example of this can be seen when the system is under a heavy disk load (in this case, running WinZip on a large AVI file). While running this activity in the background, see the following degradation when running Heroes of Might and Magic using the same methodology as in Test 2:
(See Figure 5)
Clearly, running two disks in a RAID-0 array does indeed lessen the hit that the game performance takes when running multiple tasks versus a single-disk solution. The improvement of approximately 15% is significant, but probably not enough to justify the purchase of a second disk and hardware RAID controller.
A far more dramatic picture emerges when repeating the above test using the "Separation" mode of disk configuration:
(See Figure 6)
A noticeable difference indeed! Simply by installing two separate hard drives in a system and abandoning all thoughts of a RAID configuration, it is possible for a gamer to improve their performance under a heavy load by 40% or more. In addition, this performance boost can be accomplished without incurring the additional cost of a RAID controller.
In the age of competitive gaming, people have become accustomed to being subjected to a barrage of benchmarks and buzzwords by hardware companies eager to increase revenue. It is, obviously, far more impressive to say “I have RAID-0 in my gaming rig” than it is to have two separate disks without a nice title attached – but you can see that there is a compelling reason to do exactly that for many of us.
The sweet spot of a RAID-0 configuration is, as illustrated in the first benchmark of this article, when large files need to be loaded sequentially. Examples of this would be in video capturing and editing, where massive amounts of I/O bandwidth are required to keep pace with devices dumping their data onto the bus as rapidly as possible. For gamers, however, RAID-0 offers no performance benefits – and doubles the possibility of data loss due to a failed disk.
To summarize – consider what you do on a day to day basis with your computer when deciding what hardware to purchase. Plan the configuration of your hardware to optimize your experience, not to increase bragging rights in your message board signature. If you are like most of us, and you love your BitTorrent and video, seriously consider moving those functions to a second drive to take the I/O load off of your gaming drive, and leave the marketing buzzwords to the uninformed masses.
You’ll be glad you did.
|37 User Comment(s) • 16 root comment(s)|
| MrWizard6600 (25) Apr 01, 2007 - 04:11 am|
|» you confused the hell outta me...|
So, I read your artical, it confused the hell outta me. I thought maybe my facts were off so i went to good ol' wiki. I thought I knew alot about raid, but it all turned out to be ALL wrong :S. And I've been sellin PCs for 3 years now... thats not good.
Anyways, I revert myself back to raid n00b.
but wouldn't raid 1 be the best configuration for games? I mean, even if it was as simple as, read requests 1,3,5,7,9... are sent to Hard Drive A, and read requests 2,4,6,8,10... are sent to Hard Drive B, surly you would see a significant reduction in load and caching times? no?
anywho, I would have liked to have seen how Raid-1 performance stacked up against what you were doing.
On to google I Guess...
anyways, well writting, would have liked to have seen this make the finals.
» Login to reply to this
» Note: You need to be logged in to write a comment!Login here, or if you don't have an account with FiringSquad, register here, it's FREE!
My Media-Blog categories