Raid 0 or Not?

[gthorn2000]

OK heres the deal. I have two 120GB Seagate Barracuda 7200.7 SATA drives and I have been using Raid 0 with a 128k stripe size for close to two years with no stability problems. However, my PC needs a fresh start for the summer so I am thinking about adjusting my HD setup for better performance. I want the system to be optimized for gaming, but I do occassionally do some pretty hardcore video editing. MY QUESTION IS:

1. Should I use Raid 0 or Not?

2. If so, what stripe size should I use?

3. Also, what is the best way to partition the drives?

I would appreciate it if someone would post some benchmarks or tests that prove their statements one way or the other.

 

[RavenD]

Dont use raid-0.
Just use the two drives separately. Something along the lines of one for windows/games/apps, the other for storage and pagefile.

 

[lithium726]

I use RAID0 just cause i want a 500GB drive and i dont want to pay for it... so two 250's do the trick. I would suggest getting a new drive if youre looking for more performance... those 300gb drives will not only be faster than a raid0 stripe of two 120gb drives, it will hold more and only cost around 100 bucks.

 

[drizzt81]

I would appreciate it if someone would post some benchmarks or tests that prove their statements one way or the other.

with us not having these drives available that will be difficult. Also, we all know that a lot of benchmarks do not neccesarily reflect real-world performance.

 

[Jason711]

raid 0 is not worth the VERY LITTLE it gives you.

 

[gthorn2000]

Well since I am content with 250 GB of storage, I don't plan on purchasing new HDs any time soon. Does everyone agree that it is best to not use Raid 0 in favor of putting OS, Apps, and Games on one drive and Pagefile, storage on the other?

 

[Jason711]

if you want to see a single partition, you can mount one drive to a directory on the other drive.

 

[Stugots]

I use RAID 0 because its when I work with DVD/CD images it runs alot faster than with a single drive.

 

[NebulaD]

I use RAID 0 because its when I work with DVD/CD images it runs alot faster than with a single drive.

agreed

i have also been running raid 0 for 2 years now and I have the same exact drives you do. 2 120 gb sata drives. I have no had a problem with it yet.

 

[Jason711]

large file transfers it will help.

 

[gthorn2000]

What is the best stripe size for gaming?

 

[drizzt81]

I use RAID 0 because its when I work with DVD/CD images it runs alot faster than with a single drive.

that is almost a given, but does it run faster than two independent drives? One drive could be the source, the other the destination for a 'bulk edit' and you could literally work on the bytestream...

 

[unhappy_mage]

that is almost a given, but does it run faster than two independent drives?

Answer: No. I tested this. Copying from A to B is much faster than duplicate from A+B to A+B. That's essentially what most CD/DVD operations are - creating ISO files is easy enough on the CPU that it's all about byte-copy speed.

 

[UltimaParadox]

Maybe a bit older now, but I think its still relevant, and they used two 74GB Raptors for their tests

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

Quick review

If you haven't gotten the hint by now, we'll spell it out for you: there is no place, and no need for a RAID-0 array on a desktop computer. The real world performance increases are negligible at best and the reduction in reliability, thanks to a halving of the mean time between failure, makes RAID-0 far from worth it on the desktop.

There are some exceptions, especially if you are running a particular application that itself benefits considerably from a striped array, and obviously, our comments do not apply to server-class IO of any sort. But for the vast majority of desktop users and gamers alike, save your money and stay away from RAID-0.

Hope that helps

 

[piako]

Maybe a bit older now, but I think its still relevant, and they used two 74GB Raptors for their tests

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

Quick review


Hope that helps

i completely disagree with that, ive been running a 1 TB raid 0 primary boot array over 4 disks for about two years and it is much faster than a single drive. game loading times are nil. i hate how these sites bash raid-0. you need to know the risk i agree considering you increase your chance of array failure with each added disk. people don't really consider how raid-0 is not much different than a single disk because if your single disk fails where's your backup? raid 1 is fine but it's slow and wastes resources. What people want is essentially a secondary storage system that is fast and never will fail. the problem isn't the speed it's the risk. raid 0 = risk. single disk = risk. raid 1 = less risk but is slow based on testing i've done. just gotta pick what you think is best for your needs and expected down time worth if the array fails. i've not had a problem with this array at all. eventually it will begin to fail i realize that but life's short and i'd rather have blazing speed and high risk than waiting for the parity calculations to compute. waste of time imo because i don't want to be burried knowing i wasted time for a half arse setup.

perhaps the os x raid sw is better than pc versions that's why i have better results than others in terms of speed.

i've found 16k to be the fastest stripe size on a number of different raid 0 test setups ranging from 2-4 disks over 1-2 different controllers. 16k best

 

[unhappy_mage]

i completely disagree with that, ive been running a 1 TB raid 0 primary boot array over 4 disks for about two years and it is much faster than a single drive. game loading times are nil.

Okay. Let's see your benchmarks. Especially the ones where you compare your setup to a single Raptor 150; I'd love to see those.

None of the benchmarks I've ever seen you post take into account seek times. If you're running OSX I can't recommend anything. But I assure you they're pretty bad compared to a single disk, especially compared to a single Raptor.

The reason raid 0 is "bashed" so much is there are more reliable, cheaper, faster solutions. A single Raptor is faster than a raid 0 setup for most things, and more reliable - there's only a single point of failure, instead of two. If you haven't done the math on how reliable your array is, now is a good time to do so.

And lastly, using one large array for everything is a bad idea. Get a seperate boot disk and data storage array, even if you must make it raid 0. Seek times will be improved by this simple change - if the boot files are more localized, the seeks are shorter, and take less time to get done.

 

[mwarps]

Answer: No. I tested this. Copying from A to B is much faster than duplicate from A+B to A+B. That's essentially what most CD/DVD operations are - creating ISO files is easy enough on the CPU that it's all about byte-copy speed.

Okay, I'm a little confused by this. Are you saying that a raid0 of two decently fast drives is going to be slower than a single spindle 10K rpm drive for an operation such as DVD ripping? (copying the entire DVD to the hard drive, for example)

I am going to have to test this out when I get a chance and pay a little more attention, because my sister's machine has a raid0 of 2 WD 250s, which she uses as an editing scratch drive. When I was building her machine, I stole it for a few days to do some West Wing rips, and I compared it to my own machine, which had the same proc/ram, but was using a single spindle 120GB WD drive, and it was much much much slower.

Read speed with the raid was about 26MB/sec and with the 120 it was about 18MB/sec using virtualdub. Not the most accurate or appropriate benchmark, but it was real life.

 

[unhappy_mage]

Okay, I'm a little confused by this. Are you saying that a raid0 of two decently fast drives is going to be slower than a single spindle 10K rpm drive for an operation such as DVD ripping? (copying the entire DVD to the hard drive, for example)

No, I was saying that using two disks seperately for operations which read from one file and write to another, and are disk-bound, is faster than using two disks in raid 0 for both files.

I am going to have to test this out when I get a chance and pay a little more attention, because my sister's machine has a raid0 of 2 WD 250s, which she uses as an editing scratch drive. When I was building her machine, I stole it for a few days to do some West Wing rips, and I compared it to my own machine, which had the same proc/ram, but was using a single spindle 120GB WD drive, and it was much much much slower.

If the editing is mostly disk-bound (cutting commercials, then copying the bitstream) then two seperate disks would be better - put the source on one drive and the destination on the other.

Read speed with the raid was about 26MB/sec and with the 120 it was about 18MB/sec using virtualdub. Not the most accurate or appropriate benchmark, but it was real life.

VD does some strange things while reading files. For example, here's a trace of it loading a VOB file:

1:15:24 PM	VirtualDubMod.e:2692	READ 	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753664000 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753664000 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ 	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753729536 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753729536 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ 	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753795072 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753795072 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ 	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753860608 Length: 65536	
1:15:24 PM	VirtualDubMod.e:2692	READ	\\fs\data\Movies\2001 A Space Odyssey.vob	SUCCESS	Offset: 753860608 Length: 65536

It's reading off a network drive, which isn't the best benchmark, of course, but it's reading every block twice. That's a little odd, but it shouldn't hurt performance too much. I got 174 of these requests (so actually 5ish megs of data) done in a second. The network utilization is about 70%, so I think it can only be reading the blocks once, and they're cached for the second request.

Anyways. I'd suggest that you try two disks, and do A->B and then B->A. That should help speeds a little bit.

 

[drizzt81]

i completely disagree with that, ive been running a 1 TB raid 0 primary boot array over 4 disks for about two years and it is much faster than a single drive. game loading times are nil. i hate how these sites bash raid-0. you need to know the risk i agree considering you increase your chance of array failure with each added disk. people don't really consider how raid-0 is not much different than a single disk because if your single disk fails where's your backup?

RAID-N (where N is an integer > 0) is not backup.

 

[gthorn2000]

So ultimately the best situation is two have 2 disks that aren't in a raid array?

 

[MrGuvernment]

raid 1 is fine but it's slow and wastes resources.

Slow? ever seen the read speeds from a raid 1 array - they arent slow, writes to a raid 1 is slow, but not by SO much it is SLOWWWWWWWW - beside the extra seconds a file may take i would rather have knowing if anything dies - i got a back up right there, up to date! ready to go.

And wastes resources? Please specifity...... with RAID 0 you waste resources and more time burning backups to DVD or CD, or copying them to another harddirve......

 

[mwarps]

Slow? ever seen the read speeds from a raid 1 array - they arent slow, writes to a raid 1 is slow, but not by SO much it is SLOWWWWWWWW - beside the extra seconds a file may take i would rather have knowing if anything dies - i got a back up right there, up to date! ready to go.

See three posts up. raid is not a backup.

 

[MrGuvernment]

See two posts up. raid is not a backup.

No it is not a perfect back up, but it is a partial one that is better then no back up. I guess the word "redundancy" should be used instead of backup.

What defines a backup can go on for life! for some a DVD is a backup - others a raid 1 is a backup since you have 2 copies of one file and if one fails, you have the other to fall back on.

Others having DVD / HD and off site storage and a tape drive is backup - what ever suits your needs.

All mediums used for backing up can be faulty or fail.

 

[unhappy_mage]

All mediums used for backing up can be faulty or fail.

But not all backups are automatically corrupted when the source goes. When you delete or modify a file from a raid 1 array, it's gone. When you delete or modify a file from a system that's backed up to DVD, the DVDs don't change. This is the difference between backup and redundancy.

 

[piako]

the online database of benchmark results i use is down. so i can't get the raptor results when i used to run a v1 7400gd and compare them to the raid setup now. but this page has some good info

http://www.barefeats.com/hard38.html

once the xbench database is up i'll put the raid0 and the raptor side by side so you all can see the speeds.

 

[drizzt81]

the online database of benchmark results i use is down. so i can't get the raptor results when i used to run a v1 7400gd and compare them to the raid setup now. but this page has some good info

http://www.barefeats.com/hard38.html

once the xbench database is up i'll put the raid0 and the raptor side by side so you all can see the speeds.

So how exactly is the test that you show a 'real-world' test? How often do you duplicate a 475MB file? Why does it not show us "file duplicate A to B" compared to the RAID? How frequently do 'real-world' applications read or write 100MB chunks instead of just a couple of KB? How exactly does the unnamed benchmark deal with seeks?

 

[unhappy_mage]

the online database of benchmark results i use is down. so i can't get the raptor results when i used to run a v1 7400gd and compare them to the raid setup now. but this page has some good info

http://www.barefeats.com/hard38.html

once the xbench database is up i'll put the raid0 and the raptor side by side so you all can see the speeds.

All of those benchmarks are long, linear reads and writes. If you're not using a benchmarking application or timing a real-world one, your results mean nothing with regards to the real world. Let's see them get out the IPEAK and trace some real-world applications like Femme Taken does. I leave it to you to determine what his results show and why.

 

[piako]

mage that femme taken link is nice thanks for that and your constructive comments. i've found that activity monitor will post the IOps and duplicated a 25 gig file. The IOps or IO/sec came out to average about 550 and peaked once at 1024 and peaks at 755 in this graph where the green spike is. for some reason activity monitor does not give you the average but will retain the peak IO where the 755 is displayed and does not seem to update to display what i would like, the average iops. The femme taken site seems to show that the raid 5, 0, and 10 arrays perform very similar with raid 0 taking the cake on most tests. the raid 5 arrays seem to come close on those insane scsi setups. i agree that the duplicate file thing on the barefeats site is kinda lame. honestly there are really not many benchmark apps out there for os x compared to the many posted on femme taken's site for windows. the best i've found so far is the duplicate file and then read the iops on the monitor program. ideally i think what we want to see is the maximum array performance in iops to give some comparison to the femme taken site. i see they have

a single raptor giving 346 iops
two 36 gig raptors in raid0 giving 490 iops
my array around 500 iops over 4 wd2500ks in raid0

it's a little confusing that the RAID Desktop StorageMark 2003 software has different load packages for heavy and light workstation etc. it's hard to tell which specific task or load package best equates to the duplicate file operation i performed to measure the raid setup on my machine. because of this the results are off i guess. it's a bit frustrating that there isn't a cross platform test to universalize results. as for 'everyday use' it's up in the air i guess. i listen to aiff cd audio (not mp3 or the like compresed) while capturing video and audio off of program streams while working on the internet and working in gimp and a bunch of other apps all at the same time. i don't know if the raid0 really helps or not to facilitate all that data. i do know tho that the raid volume helps with video capture since that is a continuous write. like i said i'm able to capture 30 fps on h.263 (real time compression) while listening to uncompressed cd audio in itunes, working on the internet, openoffice, gimp for images at all the same time so there are a number of different platter locations being written to and read from. if i didn't have the raid array the capture would drop frames, itunes would get a buffer underrun or some process would bring the system to a halt. i'd prefer to run raid 5 for safety but no controller for os x allows you to boot raid5. the firmtek cards only allow you to boot raid 0 and 1 and jbod (which isn't really raid imo). my situation is that the raid 5 controller are like uber expensive for mac and require special software formatting. perhaps i'll buy a rocketraid 2220 and put all the video and stuff on a raid 5 volume and boot off of raid0. either way i would really perfer to boot of a raid5 or raid 10 volume but that's not an option. i'm thinking of moving to unix since i'm tired of paying for os software and once 10.5 comes out it'll be another $100 for that. perhaps i can boot off of raid 5 or 10 in unix? any ideas? thx again

 

[Jason711]

oops.. sorry.

 

[MrGuvernment]

But not all backups are automatically corrupted when the source goes. When you delete or modify a file from a raid 1 array, it's gone. When you delete or modify a file from a system that's backed up to DVD, the DVDs don't change. This is the difference between backup and redundancy.

Thanks for the clarification

 

[unhappy_mage]

it's a little confusing that the RAID Desktop StorageMark 2003 software has different load packages for heavy and light workstation etc. it's hard to tell which specific task or load package best equates to the duplicate file operation i performed to measure the raid setup on my machine.

IOPS is not a global constant. There's a difficulty value associated with each test, if you will, making it harder or easier to get a certain number of IOs done per second. So like you said, there's no test that corresponds to your test.

like i said i'm able to capture 30 fps on h.263 (real time compression) while listening to uncompressed cd audio in itunes, working on the internet, openoffice, gimp for images at all the same time so there are a number of different platter locations being written to and read from.

The video compression is almost certainly cpu dependent. What bitrate are you compressing to?

perhaps i'll buy a rocketraid 2220 and put all the video and stuff on a raid 5 volume and boot off of raid0.

I'd suggest you look into having a single disk for a boot disk, perhaps a raptor 150, and a raid 5 to store everything on. Get hardware raid like an Areca if speed is a priority. Raid 0 is just too flaky in my experience.

either way i would really perfer to boot of a raid5 or raid 10 volume but that's not an option. i'm thinking of moving to unix since i'm tired of paying for os software and once 10.5 comes out it'll be another $100 for that. perhaps i can boot off of raid 5 or 10 in unix? any ideas? thx again

The usual way to do this is have a small /boot partition at the beginning of the first drive, and make the rest of each drive one large partition and stripe those together in software raid of your choice. This way you have a simple, easy-to-understand partition to load the kernel from, and it knows how to get the rest of the drives online and working.

To be honest, my solution would be to abandon raid 0 as a boot disk. Maybe reinstalling is trivial for you, but although I can get the operating systems loaded pretty quickly, getting everything set the way I like it takes a few weeks.

 

[piako]

Yeah I hear you on that. It would be nice if every dektop computer came with a raid 5 array of 4, 300 GB SCSI disks (heard they are coming out soon). The iops do confuse me because if say you are measuring a database for a piece of software used across an organization people will be imputting multiple requests to the server and the heads will be moving to different locations on the platters for each different piece of information compared to a simple copy file operation with one user. Your absolutely correct on the video compression issue. I guess h.263 is max res about 320x240, so when I did a lot of measurements last night after I started trying to figure out which process uses how much cpu and how much iops the encoding was a much more cpu heavy task and used little of the bandwidth on the array considering the video isn't more than 3 MB/S. It's interesting with these new h.264 codecs because compared to the high data rate and low cpu usage motion jpeg b/a they (h.264) use much more cpu and less disk space. I actually wasn't able to capture in real time h.264. it dropped about 23 frames out of 30. it's a tradeoff. i found h.263 to give excellent quality while giving a small file size among testing i did.

when i capture the data rate it's not limited which is an option. the data rate limits don't work well tho because of the poor codec and application builds in quicktime. the needed array speed for capturing uncompressed hd is about 237 MB/S. I think the motion jpeg b captures i do are about at max 20 MB/S, if I get around to it I'll post some testing and picks of data rates for different capture modes.

I think i'm gonna be holding off on the htcp i plan to build not really because of money but partially because the setup i wanted to build would run about $3,000, but also because i think i'm going to build a really hi-fi audio station that will cost about 1/4-1/3 the cost and allow me to listen to audio. that machine may run raid 1 just so i don't have to sit around reripping the 80+ cds i have acquired over the years. thank's for the suggestion of the areca card. i've never seen anything like that with the quality and the number of ports they offer. that's really neat to see that.

as for running unix i'll prob consider moving to it if i can find applications and such and then find a way to transition or dual boot os x.

yeah i had a lot of problems when i first got into raid. my first setup was i think two wd2500jd in raid 0 for a htpc and a raptor to boot (no pun intended) . it was pretty good but the problem is mostly the sata cables that if moved like a nanometer will crash the array. there really is a problem with sata cables unseating from their disk connectors easily that adds frustration to running an array. sometimes the os will freeze or sometimes the array will become unresponsive. that's only happened to me about maybe twice over all the different setups i've run over the years. all I did to correct the problem was turn off the machine and then reseat all the power and sata connectors. this always worked to solve the problem.

i do notice considerable speed increases in booting, installing software, file navigation. raid 0 how can i say makes things more snappy. it's hard to measure this i guess but i guess if i had to say it it makes waiting for things to happen dissapear.

on kind of a side note it'll be interesting for these film studios and broadcast people to be mandated to get into hd when analog turns off in what like 2010? the data rate for 720p which i guess is used for sports and 1080p is insane and would always require some type of raid array. a lot of the work i've seen done in 1080p at the san francisco institue of digital photography (sp?) is very cpu and hdd heavy. a quad core g5 is almost needed to playback h.264 and hd even if it is compressed. my machine while being fairly fast for everything i do totally dies when i try to play back h.264 movie trailers or anything hd. it simply can't handle the cpu demand of hd. when these new hd optical drives come out for your computer it's gonna be really interesting because i bet the minimum system requirements will be like a 3 GHz p4 or a dual 2.5 GHz G5 simply because anything less will absolutely not be able to play back hd dvds or the like in full 24 or 30 fps. i've tried it on a 1.8 GHz G4 which is my main computer and i can't get it to stop dropping frames. the cpu demand again is like insane for these hd medias. a lot of $$$$$ is gonna be spent by these tv statons, film studios, to capture hd, remaster to hd in 2k or 4k, and pump out hdtv. it's almost mind numbing considering they expect to turn off analog tv. the nice thing is i think hd includes backwards compatabiltiy for regular 480p but does it do 480i for tv? if not then it's gonna be interesting to see how people react when they want to move to hd but then realize the specs they need for hd. i think i can say that most people on here use their arrays for video storage and the like (files etc.) these hd ripps are going to become much bigger in file size and take longer to rip to a disk. that's why i'm reluctant to move to a new htpc with a large array if hd is around the corner peeking at me like 'hey you're gonna be f!cked when hd hits you whilest sleeping,' because a 2TB HD array would hold much less footage compared to a regular sdtv array. it seems video is moving to much higher cpu usage for playback and both high cpu and high disk usage for mastering. I saw a breakdown of a hd dvd player the hd-1a and it had a 1 GHz proc i think, 1 GB ram. It's basically a computer now these things are. The 360 has a 3 core 3 GHz proc with a separate ati video gpu. the transitions make me think a lot about where to open my wallet in terms of buidling new raid arrays and such. the only other thing my station dies on is trying to work on 1GB+ image files in gimp (need more ram i guess or a faster proc). everything else is fine.

 

[MrGuvernment]

300 GB SCSI disks (heard they are coming out soon).

they have been out for sometime , since last year

 

[drizzt81]

a lot of the work i've seen done in 1080p at the san francisco institue of digital photography (sp?) is very cpu and hdd heavy.

So you are saying that real-time HD video editing is CPU intensive. I am sure that with the appropriate hardware implementation it will not be a problem anymore.

a quad core g5 is almost needed to playback h.264 and hd even if it is compressed.

First off, the compression would be what is making the playback so CPU intensive. Then Apple says:

Does H.264 require special hardware?
While H.264 is a computationally advanced codec, it runs on today’s shipping computers with no additional hardware required. For example, a full HD movie (1920x1080, 8 Mbps, 24 fps) encoded with H.264 plays back beautifully on a dual Power Mac G5. Internet-sized content (40kbps - 300kbps) will run on the most basic of processors, like those in mobile phones and consumer-level computers.

http://www.apple.com/quicktime/technologies/h264/faq.html
so apparently, a dual G5 will do fine for HD playback in software. By 2010, it will be a non-issue.

WAY OT!

 

[Super KW]

maybe raid 0 is not that extra preformance but what if you want more space on your C drive , the fastest 10K rpm is only 150GB but when setup raid 0 2x 150GB well get 300GB 10k rpm

 

[piako]

|

I found an application that is somewhat similar to HD Tach for OS X. It does not give the ms readings and other information, but shows in a histogram like the min, average, and max reads and writes. The first picture is 4 drives over 2 controllers. Second is 4 drives on a single controller, both in raid 0. Since the earlier posts were unclear about my unique performance so pehaps this will help future readers.

 

[unhappy_mage]

And...?

 

[lithium726]

maybe raid 0 is not that extra preformance but what if you want more space on your C drive , the fastest 10K rpm is only 150GB but when setup raid 0 2x 150GB well get 300GB 10k rpm

i guess if you really wanted to stick with sata that would be true... but there are 320GB 10k drives in existance, you know.