Advice on raid set up

[chunkymonster]

HDtach is synthetic, so it has no relevence to real-world apps. Secondly, it's a pathetically shoddy test, there's a reason few use it to benchmark HDD's.

Look, if you think you've noticed a speed increase, fine. Don't throw random numbers out and decieve people into paying for something that's going to be useless for their needs. Did you read the links I posted? RAID0 has been categorically, scientifically disproven as having ANY benefit for gaming. It's as if you're arguing that the world is flat, or that the sun orbits the earth.

Wow! I can tell you feel strongly about this. A little too strongly, I think. Relax, it's not that big of a deal. You state it's "fine" if I think I've noticed a speed increase but yet still feel the need to prove something to me. I was not, am not, and did not attempt to deceive anyone nor did I suggest that anyone run out and buy a hard drive to implement a RAID0, so you can take your self-righteous-I-gotta-prove-a-point attitude and shove it up your arse because I really don't give a flying f**k. Who are you to decide what someone else needs? If someone wants to implement a RAID0 because they think it's gonna gain them something, what the f**k do you care? Feel free to sit back with your smug face and marvel at their stupidity and know that you're better than them, and you have the links to prove it.

Once more just for fun...

Regardless, some folks here seem to get wood "proving" that RAID0 offers no benefit for gaming...makes me laugh...whatever... :roll:

Got wood?

 

[cb62fcni]

Actually, I do. Talk dirty some more.

 

[SuperFly03]

Actually, I do. Talk dirty some more.

Woah, what did I walk in on? 8O

lol.

 

[Red_Frog]

Really, 20%? Wow, I must have done my benchmarking wrong. But oh wait....

http://www.anandtech.com/storage/showdoc.aspx?i=2101&p=11

http://www.bjorn3d.com/read.php?cID=734&pageID=1211

http://www.overclockers.com/articles1063/index02.asp

I guess I was right after all.

Look, RAID0 only matters in synthetic and professional apps, there are no ifs, ands, or buts about it. It occassionally even INCREASES load times in games due to greater seek times!!!! I didn't spend hours of my time running benchmarks because it makes me happy to disprove RAID's effectiveness. I did it because of people that pull random numbers out of their butts and wind up costing people good money that could be better used elsewhere. I just hope y'all aren't too thick to get it.

Did you even bother to read the articles you posted? Raptors are, for whatever reason (I don't have a clue, you would have to ask WD), notoriously crap for RAID. At least from my own experience, which is not ironically what I am most interested in.

Also, the second link you posted did nothing apart from raise further questions regarding your illiteracy.

 

[cb62fcni]

You mean literacy, right? No, seriously, I posted the second link to show the gains you get using raid for synthetic/professional apps as opposed to gaming. And also to show that they're almost never 20% as some claimed.

The Raptors are far from crappy for RAID, I don't know where you got that information from. In some cases they're even faster than 15k SCSI's in the same apps:

http://www.hardwarezone.com/articles/view.php?id=690&cid=10&pg=9

The key thing is that they have a very fast I/O.

 

[ShadowdogKGB]

Anandtech did Far Cry and Unreal benchmarks. Two old games. WOW. JUST WOW. Overclockers did 5 games and 4 synthetics. Oh and Overclockers used a stopwatch so let's not even go there. You see a 1-2 second deficit on some of Bjorn3d's PCmark04 tests and people start preaching the evils of Raid 0...

Here's a more balanced thorough analysis. http://tweakers.net/reviews/515

Motorhead I'll tell you with out a doubt you're going to see insane load times with 4 of those drives in raid 0. It ain't perfect but jeesus what is?

 

[cb62fcni]

Ok, so what does this tweakers.net (heh heh heh) article prove that I haven't already stated? Once again, just like in the bjorn3d post I put up, it shows RAID's benefits in synthetic apps. I never said RAID had any problems with synthetic stuff, in fact, for photoshoppers and video editors it's great. Where's the games?

And I was rolling laughing looking at their "subjective" results. I'll have to incorporate that into my next psychology disseration.

 

[choirbass]

not to retract any of my statements before... but, i found out that, i guess a lot of raid performance in general, is dependant on the specific controller in use (thats probably obvious)... but, that onboard raid in itself, basically sucks, compared to 'real', dedicated raid controllers (even some pci cards that you would think are dedicated, really arent, and are on par more with onboard), onboard being unable to intelligently organize data requests from what ive heard (and onboard possibly being unable to really help load times at all, which is another very real possibility)... but also, i really cant help one way or the other now, until i know more... ...i still stand by what i said before, but theres another major factor taken into account too... that might potentially throw off a lot of what i said, or maybe help improve upon it... so, i wont say the data was wrong, but, i will say it was incomplete... a great deal of performance i guess also just depends on the type of controller in use (not to mention stripe sizes, cluster sizes, etc).

in the end, you may be better off going with raid 0 for gaming, and you might very well not, possibly greatly depending on your hardware configuration.

 

[cb62fcni]

Yea, the controller matters a lot, one of the best controllers on the market today is the SB600 on CF3200 boards, the ICH8R is no slacker either.

However, no matter how efficiently the onboard controller stacks data reqs, they're still bottlenecked by HD seek times when trying to move bits and pieces about. WD Raid editions are one example of manufacturers trying to streamline I/O to overcome this somewhat. These products are intended for server/professional applications, gaming performance will not be improved tremendously no matter what controller/HDD is used.

Hybrid flash drives should in the future change the situation entirely. Because flash is operating with effectively 0 latency, HDD seek times will no longer be much a part of the equation. In theory, you should see diminishing returns on the order of 33.3% for each drive you add to RAID0, so long as the accessed file is residing in the flash RAM. So with even a 2 drive setup and stripes bigger than 128K, game load times and other similar tasks should see quite a reduction. We'll have to wait and see how well exactly this pans out, but I'm hopeful. By next year there should be benchmarks on this, considering these drives are currently entering the market.

 

[choirbass]

it seems those are onboard, and performance might be on par as such... but, im curious how they would compare to something like an Areca ARC-1210

http://www.newegg.com/Product/Product.asp?Item=N82E16816131003

i was told about that controller in particular by someone, which i then googled for... but, i 'guess' that controller in itself would blow away any onboard raid solution (im guessing here, because i really dont know... but it does sound that way)

but, supposedly thats what would be considered a true raid controller, with the performance to back it up... most onboard raid controllers only offer a small % performance improvement being in raid as weve seen, but controllers similar to the one i linked to above, supposedly offer up to near 100% improvement (or reasonably close anyhow)... he didnt say 100% improvement that i know of, but, it wouldnt be entirely unrealistic to see either

 

[cb62fcni]

I'm willing to bet it would speed up RAID performance substantially. while at the same time reducing the processor overhead inherent in running an onboard raid solution. Good find, I'd love to see a review of it.

 

[Red_Frog]

You mean literacy, right? No, seriously, I posted the second link to show the gains you get using raid for synthetic/professional apps as opposed to gaming. And also to show that they're almost never 20% as some claimed.

The Raptors are far from crappy for RAID, I don't know where you got that information from. In some cases they're even faster than 15k SCSI's in the same apps:

http://www.hardwarezone.com/articles/view.php?id=690&cid=10&pg=9

The key thing is that they have a very fast I/O.

No, that wasn't a typo. If I may quote a line from the "conclusion" of the second link, "...I would definitely recommend taking advantage of the native NCQ capability of their system by pairing it with an NCQ-capable drive or, better yet, multiple drives configured in a RAID array."
Which I might also point out, is referring to the onboard RAID of the 915 chipset. Only reason I mention this is because choir brought up the point that onboard RAID controllers are entirely capable of being fluff.


Also, I suppose you have first hand experience with RAID in both Raptor and 15k SCSI environments? Not just links to your reputable sites?
The reason I ask, not to be a prick, is that there's really no discussion involved if you're simply rehashing someone else's information.

 

[cb62fcni]

You seem to be misunderstanding me. My argument isn't that RAID is always bad, far from it, it's just not much benefit to gaming. The whole point of the bjorn article was to show its strengths in synthetic apps, which are somewhat modest, but definately worth it. I put it in because someone had tried to cast me as some sort of anti-RAID crusader, but I was only trying to discourage people from dropping RAID into a rig solely for gaming, when the money could be much better spent of RAM/CPU/GPU.

As a matter of fact, I've worked quite a bit with SCSI, and I've operated single Raptors in 3 of my computers. Never had the raptors in RAID though, that was the first time I'd heard that they were crap at it, so I did research. I try not to post "1st hand" info a lot, because too many people will just say anything to prove a point, if its in an article from a reputable source it's just more believable.

 

[skligmund]

I enjoy playing Day of Defeat Source (a Half-Life 2 Mod). I originally had one Seagate 7200.7 80GB SATA HD. My map load times were consistant with everybody else. I would show up in map maybe 8th or 9th on a 24 man server (actually, my clan's server). With no changes, other than the addition of an identical 80GB Seagate SATA HD, I have yet to be beat into the map (unless I have to download things). I might add, that since I live in Alaska, my ping is around 100 average, vs their 5-50. I also have only 768 DSL. So my internet isn't the reason. I noticed the speed increase from a RAID0 setup.

If you believe that you do not get a loading performance boost from a hard drive intensive load, you obviously have done something wrong.

In another game, I have yet to try with only one drive, Flight Simulator X load times are significantly shorter than those with comparable (same RAM/CPU and game settings) systems without RAID. No, I don't have numbers for you, regardless it is so.

I have EXPERIENCED this speed increase, and I KNOW it is happening. I know it well enough that I have 2 more identical drives in the mail currently. I will benchmark the differences between:

RAID0 (2 drives)
RAID0 (4 drives)

And anything else I might try, depending my results of the first 2. I'm not going to do just to do it, i want to see if there are benefits. If the RAID0 across 4 drives isn't significantly faster than the 2 drive setup, then I may play with the other number RAIDs, and see what they do.

 

[fidgewinkle]

Really, 20%? Wow, I must have done my benchmarking wrong. But oh wait....

http://www.anandtech.com/storage/showdoc.aspx?i=2101&p=11

http://www.bjorn3d.com/read.php?cID=734&pageID=1211

http://www.overclockers.com/articles1063/index02.asp

I guess I was right after all.

Look, RAID0 only matters in synthetic and professional apps, there are no ifs, ands, or buts about it. It occassionally even INCREASES load times in games due to greater seek times!!!! I didn't spend hours of my time running benchmarks because it makes me happy to disprove RAID's effectiveness. I did it because of people that pull random numbers out of their butts and wind up costing people good money that could be better used elsewhere. I just hope y'all aren't too thick to get it.

One of these tests, the one done by bjorn3d, shows RAID0 to give a noticeable performance improvement. I'm not sure why you didn't notice.

Now, I'm going to tell you a credible reason why two of these tests probably failed. They were running multiple Raptors on old P4 MOBOs with SATA I controllers. The maximum throughput on SATA I is not much greater than the maximum throughput of a Raptor, reducing the value of having a RAID0 configuration. It seems that the increased access time offsets the small data transfer rate improvement in the tests run by Anandtech and Overclockers. This does not mean RAID0 is not viable. It just means a RAID0 array of Raptors on a SATA I controller is not effective. For this to be a rigorous scientific inquiry, it would behoove us to look at more than one flawed hardware configuration.

In support of my assertion, the Bjorn3d test that was successful used two 7200rpm drives, which have a lot more headroom on SATA I than the raptors do and can get closer to double throughput when data is being transferred. Clearly, it isn't the case that RAID0 is useless, and it seems that there is some value to RAID0, even in games.

 

[choirbass]

well... if theyre on sata, theres essentially no ceiling limit (for all practical purposes)... the ceiling limit of sata150, is 150MB/s 'per drive', same with sata300, which is 300MB/s 'per drive' (but no single hdd is going to saturate either interface, not even pata/100 for STRs, for a single hdd anyhow)... if you have 4 raptors in raid 0 for instance, youre allowed up to 600MB/s total (not a max of 150MB/s)... i havent referred to the articles, but the performance, would greatly depend on the controllers used, since there seems to be such disparity in performance... onboard controllers simply suck when it comes to raid 0. simple as that.

 

[Flakes]

well... if theyre on sata, theres essentially no ceiling limit (for all practical purposes)... the ceiling limit of sata150, is 150MB/s 'per drive', same with sata300, which is 300MB/s 'per drive' (but no single hdd is going to saturate either interface, not even pata/100 for STRs, for a single hdd anyhow)... if you have 4 raptors in raid 0 for instance, youre allowed up to 600MB/s total (not a max of 150MB/s)... i havent referred to the articles, but the performance, would greatly depend on the controllers used, since there seems to be such disparity in performance... onboard controllers simply suck when it comes to raid 0. simple as that.

depends on the controller. im sure theres a controller out there that has a total of 150 then shares that between HDDs as they are added.... not sure though could of just been a rumor.(it was a long time ago when this onboard raid and sata had just started)

 

[choirbass]

each hdd on a sata controller is on its own indepentant channel, as a master hdd... now, if its onboard, there really is no shared bandwidth limitation, and each hdd is free to use up the full independant channel bandwidth (150MB/s or 300MB/s, but again, no single hdd will fill up even sata150 for STRs)... ...but if its a sata controller over standard 33MHz pci for example, no matter how many sata ports you have, youre limited to 127MB/s max, shared between all the devices on the pci bus