andsol Wrote:
-------------------------------------------------------
> some performance data ... trimmed from post
> Processor goes up to 50%, and results are better on small data.
I have NSA-320 w/ 2 3TB WD REDs in RAID1 booted from USB stick and am on 4.4 kernel and Jessie here is what I'm seeing.
I realize this isn't apples to apples for your 1 drive NSA-325v2 but maybe this will give you some data point you were looking for.
Stock setting:
The initial tests ran slow, but the follow up ones were faster.
I never noticed CPU going above 70% during this... (watching TOP for this so could be off a bit)
Here is the 'dd' testing:
Swapped out to a 1M block size for a 1Gb file, was slightly slower (60MB/s vs 75MB/s)
Reads were pretty nice:
## changed queue scheduler for drives
Performance is up, also noted CPU getting to mid 90s now (watching TOP for this so could be off a bit)
Also seems to have smoothed out / potentially improved dd testing:
Some of this could be standard deviation though.
-------------
Moderator edit: made this its own thread.
-------------------------------------------------------
> some performance data ... trimmed from post
> Processor goes up to 50%, and results are better on small data.
I have NSA-320 w/ 2 3TB WD REDs in RAID1 booted from USB stick and am on 4.4 kernel and Jessie here is what I'm seeing.
I realize this isn't apples to apples for your 1 drive NSA-325v2 but maybe this will give you some data point you were looking for.
Stock setting:
IQ2700i# cat /sys/block/sda/queue/scheduler noop deadline [cfq] IQ2700i# cat /sys/block/sdb/queue/scheduler noop deadline [cfq] IQ2700i# cat /sys/block/md0/queue/scheduler none
The initial tests ran slow, but the follow up ones were faster.
I never noticed CPU going above 70% during this... (watching TOP for this so could be off a bit)
IQ2700i# hdparm -Tt /dev/sda /dev/sda: Timing cached reads: 578 MB in 2.00 seconds = 288.75 MB/sec Timing buffered disk reads: 216 MB in 3.02 seconds = 71.53 MB/sec IQ2700i# hdparm -Tt /dev/sdb /dev/sdb: Timing cached reads: 576 MB in 2.00 seconds = 287.58 MB/sec Timing buffered disk reads: 312 MB in 3.01 seconds = 103.62 MB/sec IQ2700i# hdparm -Tt /dev/sda /dev/sda: Timing cached reads: 388 MB in 2.00 seconds = 194.00 MB/sec <====== this one was 100MB/s slower for some reason compared to the rest. Timing buffered disk reads: 332 MB in 3.02 seconds = 109.79 MB/sec IQ2700i# hdparm -Tt /dev/sdb /dev/sdb: Timing cached reads: 574 MB in 2.00 seconds = 286.39 MB/sec Timing buffered disk reads: 362 MB in 3.01 seconds = 120.39 MB/sec IQ2700i# hdparm -Tt /dev/md0 /dev/md0: Timing cached reads: 556 MB in 2.00 seconds = 278.03 MB/sec Timing buffered disk reads: 382 MB in 3.01 seconds = 126.97 MB/sec IQ2700i# hdparm -Tt /dev/md0 /dev/md0: Timing cached reads: 544 MB in 2.00 seconds = 271.42 MB/sec Timing buffered disk reads: 384 MB in 3.02 seconds = 127.11 MB/sec
Here is the 'dd' testing:
IQ2700i# time sh -c "dd if=/dev/zero of=testfile bs=100k count=1k && sync" 104857600 bytes (105 MB) copied, 1.39 s, 75.4 MB/s 0.00s user 0.91s system 49% cpu 1.829 total
Swapped out to a 1M block size for a 1Gb file, was slightly slower (60MB/s vs 75MB/s)
IQ2700i# time sh -c "dd if=/dev/zero of=testfile bs=1M count=1k && sync" 1073741824 bytes (1.1 GB) copied, 17.9524 s, 59.8 MB/s 0.03s user 11.77s system 64% cpu 18.186 total
Reads were pretty nice:
IQ2700i# time sh -c "dd if=testfile of=/dev/null bs=1M" 1073741824 bytes (1.1 GB) copied, 9.1678 s, 117 MB/s 0.01s user 7.13s system 77% cpu 9.181 total
## changed queue scheduler for drives
IQ2700i# echo deadline > /sys/block/sda/queue/scheduler IQ2700i# echo deadline > /sys/block/sdb/queue/scheduler IQ2700i# cat /sys/block/sda/queue/scheduler noop [deadline] cfq IQ2700i# cat /sys/block/sdb/queue/scheduler noop [deadline] cfq
Performance is up, also noted CPU getting to mid 90s now (watching TOP for this so could be off a bit)
IQ2700i# hdparm -Tt /dev/sda /dev/sda: Timing cached reads: 550 MB in 2.00 seconds = 275.00 MB/sec Timing buffered disk reads: 372 MB in 3.00 seconds = 123.84 MB/sec IQ2700i# hdparm -Tt /dev/sdb /dev/sdb: Timing cached reads: 538 MB in 2.01 seconds = 268.31 MB/sec Timing buffered disk reads: 408 MB in 3.00 seconds = 135.80 MB/sec IQ2700i# hdparm -Tt /dev/sda /dev/sda: Timing cached reads: 534 MB in 2.00 seconds = 266.89 MB/sec Timing buffered disk reads: 430 MB in 3.01 seconds = 142.75 MB/sec IQ2700i# hdparm -Tt /dev/sdb /dev/sdb: Timing cached reads: 556 MB in 2.01 seconds = 277.28 MB/sec Timing buffered disk reads: 416 MB in 3.01 seconds = 138.20 MB/sec IQ2700i# hdparm -Tt /dev/md0 /dev/md0: Timing cached reads: 290 MB in 2.00 seconds = 144.78 MB/sec <==== another outlier Timing buffered disk reads: 420 MB in 3.01 seconds = 139.74 MB/sec IQ2700i# hdparm -Tt /dev/md0 /dev/md0: Timing cached reads: 574 MB in 2.00 seconds = 287.02 MB/sec Timing buffered disk reads: 414 MB in 3.01 seconds = 137.58 MB/sec
Also seems to have smoothed out / potentially improved dd testing:
Some of this could be standard deviation though.
IQ2700i# time sh -c "dd if=/dev/zero of=testfile bs=100k count=1k && sync" 104857600 bytes (105 MB) copied, 1.52577 s, 68.7 MB/s 0.01s user 1.05s system 37% cpu 2.804 total IQ2700i# time sh -c "dd if=/dev/zero of=testfile bs=1M count=1k && sync" 1073741824 bytes (1.1 GB) copied, 17.3699 s, 61.8 MB/s 0.02s user 11.87s system 66% cpu 17.824 total IQ2700i# time sh -c "dd if=testfile of=/dev/null bs=1M" 1073741824 bytes (1.1 GB) copied, 8.43713 s, 127 MB/s 0.00s user 6.87s system 81% cpu 8.451 total
-------------
Moderator edit: made this its own thread.