5400 rpm hard drives, 900 MB/s

Mjolnir

Veteran XV
$zpool status
Code:
	NAME        STATE     READ WRITE CKSUM
	test         ONLINE       0     0     0
	  raidz1    ONLINE       0     0     0
	    da0     ONLINE       0     0     0
	    da1     ONLINE       0     0     0
	    da2     ONLINE       0     0     0
	    da3     ONLINE       0     0     0
	  raidz1    ONLINE       0     0     0
	    da4     ONLINE       0     0     0
	    da5     ONLINE       0     0     0
	    da6     ONLINE       0     0     0
	    da7     ONLINE       0     0     0
	logs
	  ad1       ONLINE       0     0     0


I haven't really tuned it much, but sequential writes of large files always end up 800-1000 MB/s. Read 5 different 300MB files repeatedly with 5 processes and it's just silly:
Code:
"Record size = 4 Kbytes "
"Output is in Kbytes/sec"
"  Initial write "  704429.11
"        Rewrite "  516146.90
"           Read " 3649338.56
"        Re-read " 3635120.12
"   Reverse Read " 3577227.44
"    Stride read " 3019672.62
"    Random read " 2976754.69
" Mixed workload " 2329099.28
"   Random write "   39527.96
"         Pwrite "  684943.34
"          Pread " 3785758.06


I've been involved in many RAID and storage-related threads here in the past, and I'm all too happy to bash RAID and especially RAID5 based on my bad experiences. So with performance like this, is there any argument left? I think the whole cost for this server with about 8 TB of storage (20 HDD capacity) cost about $1500.

I still haven't figured out a way to generate enough network traffic to flood this: none of the other servers will keep up.
 
Yeah for cost reasons and uhm.....lack of understanding - I'm pretty much forced to use RAID 5 for everything as well.
 
I'm confused you don't use RAID at all or just not RAID5?

I'm using raidz, which is a unique kind of software raid. It lets me combine multiple (LSI) drive controllers to make one huge pool of storage. It's a bit of a pity, but each controller isn't using any RAID features at all. In fact, each raidz set is split across multiple controllers: controller #1 hosts drive1,raidz0; drive1,raidz1; drive1,raidz2, etc. This way, if I lose an entire controller I only drop 1 drive from each raidz set, and the whole thing keeps running. It also splits up the bandwidth to keep each controller busy regardless of which raidz set is being used.


I used 5400 drives because they are cheap and because most of what I need is streaming large files, so I figured that we'd lose most of the advantages of faster drives. They also produce less heat (I'm hoping they last longer) and consume less power. The power factor is really a concern when you have 20+ drives spinning up when you turn on the server, and it's a pain to deal with staggered spin-up unless you have enterprise drives (which I don't). Noise and overall heat really weren't a concern.
 
i am so glad this thread is here.

sometime today i will be picking up 3 15K HDD's for my home computer and was wondering if i should RAID them and what kind of RAID; 5 or 10?
 
Back
Top