i8igmac Posted May 31, 2016 Share Posted May 31, 2016 (edited) this is not exactly a hack or a mod, just wanted to share benchmark results... making upgrade to my movie streaming and backup server. the motherboard is maybe 7 years old, 3gps sata ports, this might be the speed limiting factor. (currently running a live usb, will test more drives shortly) 3 cheap hard disks 5400 rpm 2 TB running raid 5. Quote mint@mint ~ $ sudo hdparm -Tt /dev/md0 /dev/md0: Timing cached reads: 6860 MB in 2.00 seconds = 3431.12 MB/sec Timing buffered disk reads: 594 MB in 3.01 seconds = 197.27 MB/sec Quote mint@mint ~ $ dd if=/dev/zero of=/mnt/raid/output bs=8k count=10k 10240+0 records in 10240+0 records out 83886080 bytes (84 MB) copied, 5.92431 s, 14.2 MB/s my old 500gig single 7200rpm Quote mint@mint ~ $ sudo hdparm -Tt /dev/sda1 /dev/sda1: Timing cached reads: 6558 MB in 2.00 seconds = 3280.10 MB/sec Timing buffered disk reads: 138 MB in 3.06 seconds = 45.05 MB/sec Quote mint@mint ~ $ dd if=/dev/zero of=/media/output bs=8k count=10k 10240+0 records in 10240+0 records out 83886080 bytes (84 MB) copied, 1.90238 s, 44.1 MB/s 1 tb 7200rpm Quote mint@mint ~ $ sudo hdparm -Tt /dev/sdb /dev/sdb: Timing cached reads: 6778 MB in 2.00 seconds = 3390.34 MB/sec Timing buffered disk reads: 358 MB in 3.01 seconds = 118.90 MB/sec Quote mint@mint ~ $ dd if=/dev/zero of=/mnt/output bs=8k count=10k; rm -f /mnt/output 10240+0 records in 10240+0 records out 83886080 bytes (84 MB) copied, 1.20226 s, 69.8 MB/s solid state drive 120gig Quote mint@mint ~ $ sudo hdparm -Tt /dev/sdd /dev/sdd: Timing cached reads: 6708 MB in 2.00 seconds = 3354.81 MB/sec Timing buffered disk reads: 440 MB in 3.00 seconds = 146.53 MB/sec Quote mint@mint ~ $ sudo dd if=/dev/zero of=/mnt/output bs=8k count=10k 10240+0 records in 10240+0 records out 83886080 bytes (84 MB) copied, 0.224919 s, 373 MB/s Edited May 31, 2016 by i8igmac Quote Link to comment Share on other sites More sharing options...
cooper Posted May 31, 2016 Share Posted May 31, 2016 The benchmarking programs you're using suck at showing actual RAID performance. Both hdparm and dd do sequential io. Admittedly, this is the expected load on a media center machine so in that respect it's fair. You should use dd for a reading test aswell to get more of an apples vs apples comparison. But let's do the math here. You have 3 drives. When you read a file from the array, let's say it's 6 blocks total, the controller (which in case of software raid is the module) will likely find on HD1 blocks 1, P3+4,6 on HD2 blocks 2,3,P5+6 and on HD3 blocks P1+2,4,5. If the controller is clever enough (=likely) it will read 1 and 6 from HD1, 2 and 3 from HD2 and 4 and 5 from HD3 and put everything in the right order. If reading any of those blocks fails, it'll read the P(arity) block for that block, subtract the block it did get and end up with the one that was missing, but in general this isn't likely to happen often. When you think about RAID you say "3 drives in an array, so should be 3x performance". Well, not quite (as you've noticed). Because of this staggering (read 2 blocks, skip the parity block, read 2 blocks, skip parity block... repeat) and the fact that the blocks need to arrive in order (so it might delay reading block 6 until 3 and 4 have been read otherwise the drive needs to jump back to read that parity block) the best you're likely to expect is a little under the max of [raid_drives-1] stand-alone drives combined. Another issue which impacts performance and which is solved by the very expensive devices is that they synchronize the disks meaning that they rotate at the exact same speed and have a 'row' (my word) of blocks, so in this example that would be blocks 1, 2 and P1+2, move under the reading arm at the exact same time. So for each rotation of the disks the controller would read or write (in your case) 3 blocks at the exact same time. This gets you closer to the maximum performance of the disk. But the bottom-line is that if you want to see the great performance that RAID is capable of, you want LOTS of disks and you want LOTS of concurrent reads with not a lot of writes. RAID wasn't developed to increase performance but to increase reliability. And as long as it delivers what you need, just be happy with that. :) Quote Link to comment Share on other sites More sharing options...
i8igmac Posted May 31, 2016 Author Share Posted May 31, 2016 I knew I made a mistake with my tool selection. I had to crack open the desktop and move all data over to the raid one drive at a time so I figured I would perform a quick test. my money ran short when I purchased these drives, bestbuy had a nice price at about 65$ per 2 tb 5400rpm. I plan to clean up this post with model numbers and I might run the test over with dd/reads/writes Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.