What Raid0 STRIPE SIZE should I choose?

[irlwizard]

My mobo has a SATA 3Gbps Intel controller, you can choose between 4 to 128kb STRIPE SIZE when setting up a raid. I will raid0 on Win7 64bit Ultimate.

What size do you guys recommend I set it to and why? I won't have ANY important data on it so I don't care at all about disk failure. It's just going to be a steam/origin gaming desktop in the livingroom and both support cloud storage so I don't have to backup anything at all... The porpose of the raid0 is to speed up load times (I will be using two new 1TB seagate barracudas), so please base the stripe size selection on that.

 

[GhislainG]

Unless your files are very small, using the default 128kb should work fine. http://www.overclock.net/t/886768/samsung-spinpoint-f4-intel-ich10r-raid-tests

 

[RetiredChief]

1) For OS and programs, Raid0 will do very little to improve. For large game maps load time should improve as LONG as the "map" is larger than the stripe size. Reason:
.. Raid0 only really improves Sequential performance. This is the least important matrix for an OS + Programs.
.. Raid0 does not decrease access time, and does Very little for the Random 4K performance.

NOTE: you can improve on access time if you use "short stoke" For a pair of WD Blacks it cut access time from about 11.+ millisec to around 9.5 millisecs. Disadvantage of short stoke is that you lose 70 -> 60% of drive capacity.

Before SSDs I used Raid0 on all my systems, going back to late 90's and using raid0 with IDE drives.

For My OS + program drive, I used the 64 K stripe size (found little advantage with smaller size) and default cluster size. For My photo and Movie drive, I went with 128 K stripe and increased cluster size to 32K.

In looking at GhislainG's link, Please ignore the comment about Raid0 approaching SSD performance. Yes you may get close on sequential read/writes but that is the ONLY real boost. Again Raid0 does NOT change access time and an SSD is on the order or one hundred times faster. As states can use short stroke to decrease, but still NO where’s approaching an SSD.

Reason Short stroke decreases access time, is that for a HDD access time is dependent on where the data is - ie is it at the outer edge of the platter, or is it closer to the center of the platter. Short stroke limits data to the outer 30->40 percent of the platter(s) where access time is highest (angular velocity).

 

[irlwizard]

1) For large game maps load time should improve as LONG as the "map" is larger than the stripe size. Reason:
.. Raid0 only really improves Sequential performance. This is the least important matrix for an OS + Programs.
.. Raid0 does not decrease access time, and does Very little for the Random 4K performance.

Most new games like Shogun 2 have HUGE load times on their maps. And a map smaller than 128kb will load instantly anyway. So I guess that is the Stripe Size I will be going with. My motherboard manual doesn't mention anything about block size. Not sure if I get to choose that at all...

Anyway SSD is not an option because they don't have 1TB+ SSDs yet (I have that many games) =)

What IS sequential reads and access time (I only look at sequential loads when in harddrive reviews)? I hear every review talking about it going bla bla bla but never actually giving the definition on what each will improve in loading. Are the WD Blacks really better than the new 1TB platter seagates, they cost almost twice as much? Does -2millisec outperform +60 sequential read/write? Oh and I actually don't belive that a raid0 won't make the OS load faster. Everytime hardwarecannucks review storage they show OS load times and worst case senario program loads and they always improve with raid by a generous amount.

Bottom line is I need 2TB of storage. Buying a 2TB will cost $100 and going Raid0 will cost $150. I don't think +50 is expensive at all... And even if I only get x1.5 the performance it will have payed itself off...

 

[irlwizard]

Best answer selected by irlwizard.

 

[RetiredChief]

1) On Sequencial read/writes.
some baseline info. HDD is setup in sectors (a sector is 512 Bytes) if you take a 1 TB drive and divide by 512 Bytes then number is HUGE, so to better manage the File Alocation table uses Clusters. a 4 K Cluser size is 8 sectors, A 10 Byte file will use 1 cluster ( the remaining 4 K- 10 Bytes is wasted ). If you write a 1 Meg file that will use 250 clusters. The system will try to write them is that one follows the last one, or sequencially. Sometimes it can not and it spreads them all over the drive (referred to as fragmentaion). When the file clusters are sequencial they can be read MUCH faster than if they are spread out randomly on the platters.

Raid0 then throws in stripe size say 64 K. For files that are >64 K parts of the files are on one drive and part is on the other drive. If it was perfect it would cut the time to read a file in half as have would be read by one drive while at the same time the other half would be read by the 2nd drive. For Files under 64 K there is no improvement in reading the file as the whole file is on one drive. Reason raid0 does not improve random 4 K performance. If you look at the files that are normally contained on a OS + program drive you will find that over half of the files are under 32 K.

With the newer HDDs you do have to be carefull in selection as many of the "consummer" HDDs should not be used in a raid0 configuration - they die at a higher rate than if used in a non raid configuration. DO NOT use any of the "green" or blue or red with yellow pokadot drives. Seagate drives seem to have a higher failure rate than WD Blacks. I have used the older 640 gig WD Blacks in raid0. Also Like the Samsung F3 drive. While it is a sata II drive it is just as fast as sata III drives. The only diff between sata II and Sata III drives is in Burst speed which is short duration. As I indicated while I loved my old raid0 systems, I quit using Raid0 when SSDs came down in price. All my systems have 2 SSDs, one dedicated for OS + Programs and the 2nd for my most used files.

You will notice an improvement with 2 HDDs in raid0 vs a single HDD.

Some advice on set-up: When you set up the raid array, DO not use the whole drive.
For example 2 x 1TB drive. Create two arrays. For the first one (drive 0) set it to use just 200->300 Gigs. This will be for your OS + Programs (use default 64 K Stripe or decrease to 32 K your choice). Then create the 2nd array using the remainder of the drive this being your "data" drive - also stick your game maps here. I'd probably go with 128 K stripe. Windows will treat this as 2 drives, drive 0 and drive 1. Install OS on Drive 0 after installing OS you can partition and formate drive 1 using windows disk manager.

 

[David Dutton]

1) For OS and programs, Raid0 will do very little to improve. For large game maps load time should improve as LONG as the "map" is larger than the stripe size. Reason:
.. Raid0 only really improves Sequential performance. This is the least important matrix for an OS + Programs.
.. Raid0 does not decrease access time, and does Very little for the Random 4K performance.

NOTE: you can improve on access time if you use "short stoke" For a pair of WD Blacks it cut access time from about 11.+ millisec to around 9.5 millisecs. Disadvantage of short stoke is that you lose 70 -> 60% of drive capacity.

Before SSDs I used Raid0 on all my systems, going back to late 90's and using raid0 with IDE drives.

For My OS + program drive, I used the 64 K stripe size (found little advantage with smaller size) and default cluster size. For My photo and Movie drive, I went with 128 K stripe and increased cluster size to 32K.

In looking at GhislainG's link, Please ignore the comment about Raid0 approaching SSD performance. Yes you may get close on sequential read/writes but that is the ONLY real boost. Again Raid0 does NOT change access time and an SSD is on the order or one hundred times faster. As states can use short stroke to decrease, but still NO where’s approaching an SSD.

Reason Short stroke decreases access time, is that for a HDD access time is dependent on where the data is - ie is it at the outer edge of the platter, or is it closer to the center of the platter. Short stroke limits data to the outer 30->40 percent of the platter(s) where access time is highest (angular velocity).

I HIGHLY disagree!

Raid0 does a WHOPPING speed for OS boot time , less OS lag time, etc. 2 Fast Sata 2 or 3 disks will give good performance.. 3 even better... up to 1.5-2X the speed if you had a single drive.

I have a 5 disk array in raid 0 and it blows away any single SSD throughput on games, file copies, etc.They go right up and smack into the 6Gps data cap of sata 3

Array enough disks in raid0, and yes, you can easily outperform a SSD.. however.. with only 2 disks, you won't.. but you will still see a nice performance boost for reads and writes for games, OS, etc.

I use a 32KB strip with a 64KB cluster size

A lot of times a file is not striped across all disks unless it's a big file... this is a GOOD thing for random reads ... one 2 drives can grab file A, another 2 file B, and the 5th a small file C.

and large files are striped across all drives...

The best way to see raid in action: Make a software raid 0 in Windows and watch the drive access...

sometimes just 1 drive is being read from... sometimes 2.. or 3.. 4 or all 5.. same with writing... and when there's a Q length of 5-10.. the drives perform incredibly fast.. I don't software raid.. I just did it to see what the drives were doing.

but my 5 disk raid 0 array can read and write at about 650MB/sec.... and Sata 3 has a cap of 6GB.. You wont see ANY DDS with 650MB read/write abilities...

Granted, I'm not posting any tech details, I'm speaking mostly in layman's terms... but I'd be happy to get technical why a large raid 0 (or even a raid 5) array can whoop the pants off a SSD

It's all about working in parallel

 

[cantisan]

Guys,

It's not clear to me if I set the strip size of 64k I need to take care with cluster size of the same time. I have SQL in Raid 10 with strip size of 64k. Should I set the cluster size to 64k too or this is not necessary ?