Raid 1 for performance...?

[wuggle]

Raid 0 = stripe = improved write + read. reduced reliability
Raid 1 = mirror = same write, improved read + reliability

Right?


So, for games and other read-heavy applications, raid 1 performance should be comprable to raid 0...? Google didnt find any raid 1 performance reviews. Am I on crack here, or did I find an unexplored section of computer hardware?

 

[imported_Kiwi]

According to testing performed and published at AnandTech, Raid 0 isn't enough improvement over normal drive storage speed to be worthwhile for any uses that don't involve the largest sizes of files. For gaming, as the example used, Raid 0 may be less efficient. And overall efficiency, if I recall the reports accurately, is also reduced by Raid 1. You get safety for your data, at the cost of slowed performance.


:thumbsdown:

 

[Matthias99]

I've seen a few benches with RAID1 performance, but mostly in server stuff, since few people run it on desktops.

In theory, it offers reduced write performance (since it has to write the data twice, rather than just once with a single drive or JBOD), but you can do two reads in parallel. Many singlethreaded applications, however, will not take advantage of this, and it does not really offer an appreciable increase in STR, since reads are normally only serviced by one disk in RAID1 (it doesn't make sense, since data is not striped as in RAID0 or RAID5). For something that does lots and lots of small parallel reads (say, a webserver), it's faster than pretty much anything else. I would not advise RAID1 for performance on the desktop.

 

[wuggle]

Originally posted by: Matthias99
I've seen a few benches with RAID1 performance, but mostly in server stuff, since few people run it on desktops.

In theory, it offers reduced write performance (since it has to write the data twice, rather than just once with a single drive or JBOD), but you can do two reads in parallel. Many singlethreaded applications, however, will not take advantage of this, and it does not really offer an appreciable increase in STR, since reads are normally only serviced by one disk in RAID1 (it doesn't make sense, since data is not striped as in RAID0 or RAID5). For something that does lots and lots of small parallel reads (say, a webserver), it's faster than pretty much anything else. I would not advise RAID1 for performance on the desktop.

In theory, raid 1 should have exactly the same write performance as no raid at all. You're writing the same amount of data to 2 hard drives. Since the writes happen in parallel, it should (in theory) take the same amount of time to write to 2 drives as it would to write to 1 drive.

IIRC, I read somewhere that the raid controller itself takes care of reading from both drives and it feeds the data to the OS faster.

I read the AT article and... well... yeah. RAID 0/1/5 doesnt offer any real world performance, but I'm just surprised that there are absolutely NO benches of RAID 1