| 1 |
Duncan posted on Fri, 28 Jun 2013 03:36:10 +0000 as excerpted: |
| 2 |
|
| 3 |
> So now I guess I send this and do some more testing of real device, now |
| 4 |
> that you've provoked my curiosity and I have the 50 GB (mostly) |
| 5 |
> pseudorandom file sitting in tmpfs already. Maybe I'll post those |
| 6 |
> results later. |
| 7 |
|
| 8 |
Well, I decided to use something rather smaller, both because I wanted to |
| 9 |
run it against my much smaller btrfs partitions on the ssd, and because |
| 10 |
the big file was taking too long for the benchmarks I wanted to do in the |
| 11 |
time I wanted to do them. |
| 12 |
|
| 13 |
I settled on a 4 GiB file. Speeds are power-of-10-based since that's |
| 14 |
what dd reports, unless otherwise stated. Sizes are power-of-2-based |
| 15 |
unless otherwise stated. This was filesystem-layer-based, not direct to |
| 16 |
device, and single I/O task, plus whatever the system might have had |
| 17 |
going on in the background. |
| 18 |
|
| 19 |
Also note that after reading the dd manpage, I added the conv=fsync |
| 20 |
parameter, hoping that gave me more accurate speed ratings due to the |
| 21 |
reducing the write-caching. |
| 22 |
|
| 23 |
SSD speeds, dual Corsair Neutron n256gp3 SATA-600 ssds, running btrfs |
| 24 |
raid1 data and metadata: |
| 25 |
|
| 26 |
To SSD: peak was upper 250s MB/s over a wide blocksize range of 1 MiB to |
| 27 |
1GiB. I believe the btrfs checksumming might lower speeds here somewhat, |
| 28 |
as it's quite lower than the rated 450 MB/s sequential write speed. |
| 29 |
|
| 30 |
From SSD: peak was lower 480s MB/s, blocksize 32 KiB to 512 KiB (smaller |
| 31 |
blocksize range but much smaller block than I expected). This is MUCH |
| 32 |
better, far closer to the 540 MB/s ratings. |
| 33 |
|
| 34 |
To/from SSD: At around 220 MB/s, peak was somewhat lower than write-only |
| 35 |
peak, as might be expected. Best-case blocksize range seemed to be 256 |
| 36 |
KiB to 2 MiB. |
| 37 |
|
| 38 |
So, best mixed-access case would seem to be a blocksize near 1 MiB. |
| 39 |
|
| 40 |
I did a few timed cps also, then did the math to confirm the dd numbers. |
| 41 |
They were close enough. |
| 42 |
|
| 43 |
Spinning rust speeds, single Seagate st9500424as, 7200rpm 2.5" 16MB |
| 44 |
buffer SATA-300 disk drive, reiserfs. Tests were done on a partition |
| 45 |
located roughly 40% thru the drive. I didn't test this one as closely |
| 46 |
and didn't do rust-to-rust tests at all, but: |
| 47 |
|
| 48 |
To rust: upper 70s MB/s, blocksize didn't seem to matter much. |
| 49 |
|
| 50 |
From rust: upper 90s MB/s, blocksize upto 4 MiB. |
| 51 |
|
| 52 |
-- |
| 53 |
Duncan - List replies preferred. No HTML msgs. |
| 54 |
"Every nonfree program has a lord, a master -- |
| 55 |
and if you use the program, he is your master." Richard Stallman |
Archives