RAID-1 vs RAID-0

[G String]

Okay, hi everyone, new here. As you know the search option is disabled, so I started a new thread.

I just have a question about RAID in general I guess.. I see so many people using RAID-0 with their disks and am wondering why.

Isn't RAID-1 better for games because the data will be read faster? (there's "twice" as much because the same image is written twice..) as opposed to RAID-0 in which the data can be written much faster?

I Googled it (http://www.m-techlaptops.com/raid.htm) and that page made it seem like RAID-0 was better for reading.

Thanks

 

[masteraleph]

RAID-0 is going to always be faster for writing, obviously.

Hypothetically RAID-1 will be roughly equivalent to RAID-0 for read implementations- RAID 1 has two copies, and therefore could read bit A off of drive 1 and bit B off of drive 2. RAID 0 on the other hand roughly alternates, so it could read bit A off of drive 1 and bit B off of drive 2- essentially the same thing, with overhead varying.

However, most (read: onboard) RAID-1 solutions do not provide anywhere near to the read benefits that RAID-1 has the potential of providing.

 

[unhappy_mage]

What masteraleph's analysis misses is the fact that drives read as fast as they spin. So when you're reading a single stream of data from the drive, you can't go any faster by reading alternate bits from alternate drives - that's only as fast as a single disk would be (not to mention that you can't transfer a single bit anyways - you need to write a whole sector of 512 bytes). What you can do to improve transfer rates is use more than one disk to service many requests. Unfortunately, many games can't even handle doing one sequence of reads well (mikeblas did some testing on FarCry, for example - I'd link it but the search is down) so expecting them to do multiple reads at once is probably a little too much to ask.

Thus, a single fast disk is probably your best bet. It'd be nice if games had better-written I/O subsystems, but in the meantime low seek latencies and high transfer rates are more useful than wishing on stars.

 

[drizzt81]

(mikeblas did some testing on FarCry, for example - I'd link it but the search is down) so expecting them to do multiple reads at once is probably a little too much to ask.

subscribed threads to-the-rescue:
http://hardforum.com/showthread.php?t=994470

 

[dandragonrage]

...you can't go any faster by reading alternate bits from alternate drives - that's only as fast as a single disk would be...

Not true for RAID-0, since RAID-0 wouldn't need to skip data to do that. It's all on one hard drive or the other.

I'm not sure about RAID-1. It's certainly possible to make a drive layout that is friendly to this, but that would be rearranging the data between drives and I'm not sure if this would still be called RAID-1. It would certainly be a variation of it if not...

You're talking about like
1-----------(skip)
(skip)-----------2
3-----------(skip)
(skip)-----------4

it doesn't have to be like that...

 

[unhappy_mage]

Not true for RAID-0, since RAID-0 wouldn't need to skip data to do that. It's all on one hard drive or the other.

Sure, that's why you get a higher STR with raid 0.

I'm not sure about RAID-1. It's certainly possible to make a drive layout that is friendly to this, but that would be rearranging the data between drives and I'm not sure if this would still be called RAID-1. It would certainly be a variation of it if not...

You're talking about like
1-----------(skip)
(skip)-----------2
3-----------(skip)
(skip)-----------4

it doesn't have to be like that...

What would it be, then? I suppose you could store the data non-contiguously, but I don't know of any manufacturer or software raid that does so.

I suppose it'd be interesting to store data like this:
disk 0: 0 2 4 6 8 10 ..... 1 3 5 7 9 11
disk 1: 1 3 5 7 9 11 ..... 0 2 4 6 8 10
That way you could get the improved sequential read performance of raid 0, with the redundancy of raid 1. The downside, of course, is that you'd always have to perform a rather large seek on both disks to perform a write to the array.

 

[dandragonrage]

I see your point about the write performance and my suggestion would be to write the data out once and then mirror it after. Combine that with large caches. But that wouldn't work well for constant IO like servers, which is where RAID is most useful in the first place... I'm sure a better placement strategy exists, but they will all have drawbacks.

 

[uOpt]

There are many more different patterns than just reading and writing.

Large contiguous reads and writes are sped up by raid-0 but not by raid-1.

But raid-1 is faster than raid-0 for example on small reads after random seeks. This pattern is sped up not at all by raid-0.

Even if you analyze all the patterns and which raid levels speed them up or not, then you are still left with the question of what access pattern exactly you have in your work. Just for example, normally application start should be sped up more my raid-1, but many games have large pakfiles that they read through efficiently, and that's faster in raid-0. Also, many modern Windoze apps write a lot of junk during startup, and that's not to raid-1's liking.

 

[Eickst]

I refer to what you call RAID-0 as "AID-0". There's no redundancy in it so why it's even considered a RAID level I will never understand.

Personally, I would forget the AID-0 and just have two seperate disks. What is the fascination with RAID on these boards?