This may be a better topic for the Illumos ZFS mailing.
There are a couple things affecting your results here.
The SSDs perform best when multiple threads are filling their queue. In
your test you have a single thread.
It is my understanding the ZIL is also one thread per ZFS file system.
The latency of the SAS bus and SSD will stack up against you here. You
should get better results across the board if you execute against several
ZFS file systems on the pool.
-Chip
On Sun, Feb 22, 2015 at 9:28 PM, Albert Chin
openindiana-disc...@mlists.thewrittenword.com wrote:
I've tested three SSDs in a Sun X4170M2. This server has 8 internal
SSD 2.5 drive bays with a Sun Storage 6 Gb SAS PCIe RAID HBA. I
believe the chipset on the HBA is a LSI SAS2108 (according to
http://tinyurl.com/koc6kdn).
I tested by adding each SSD as a ZIL for a pool and then running the
following command on one of the file systems in the pool:
$ /usr/gnu/bin/dd if=/dev/zero of=file bs=1 oflag=sync
Iostat numbers below are given by:
$ iostat -cnx 1
$ cat /kernel/drv/sd.conf
...
sd-config-list=*MACH16*,disksort:false, cache-nonvolatile:true,
*INTELSSD*,disksort:false, cache-nonvolatile:true,
physical-block-size:8192;
The 8k block size for the Intel S3700 comes from:
http://wiki.illumos.org/display/illumos/List+of+sd-config-list+entries+for+Advanced-Format+drives
Product part numbers:
1. STEC Mach16 SLC 100GB - M16CSD2-100UIU
2. STEC Mach16 MLC 200GB - M16ISD2-200UCV
3. Intel DC S3700 - SSDSC2BA200G301
ATA-STECMACH16 M-0300-93.16GB (SLC)
$ /usr/gnu/bin/dd if=/dev/zero of=file bs=1 oflag=sync
$ iostat -cnx 1
r/sw/s kr/s kw/s wait actv wsvc_t asvc_t %w %b
0.0 1439.20.0 5756.9 0.0 0.80.00.6 0 84
0.0 1478.70.0 5915.0 0.0 0.90.00.6 1 86
0.0 1491.10.0 5964.2 0.0 0.90.00.6 1 88
0.0 1506.00.0 6023.8 0.0 0.90.00.6 0 89
ATA-STECMACH16 M-0289-186.31GB (MLC)
$ /usr/gnu/bin/dd if=/dev/zero of=file bs=1 oflag=sync
$ iostat -cnx 1
r/sw/s kr/skw/s wait actv wsvc_t asvc_t %w %b
0.0 8079.70.0 32318.9 0.0 0.50.00.1 2 52
0.0 8229.10.0 32916.5 0.0 0.50.00.1 2 53
0.0 7368.00.0 29471.9 0.0 0.50.00.1 2 47
0.0 7318.00.0 29272.2 0.0 0.50.00.1 2 47
ATA-INTEL SSDSC2BA20-0270-186.31GB (MLC)
$ /usr/gnu/bin/dd if=/dev/zero of=file bs=1 oflag=sync
$ iostat -cnx 1
r/sw/s kr/skw/s wait actv wsvc_t asvc_t %w %b
0.0 9196.00.0 18392.0 0.0 0.20.00.0 1 19
0.0 9144.30.0 18288.6 0.0 0.20.00.0 1 19
0.0 9288.70.0 18575.5 0.0 0.20.00.0 1 19
0.0 8352.00.0 16704.0 0.0 0.20.00.0 1 17
The STEC Mach16's are 3.0Gbps devices. The Intel SSD DC S3700 is a
6.0Gbps device. Just two questions:
1. Why don't I see double the IOPS performance between the
6.0Gbps device than the 3.0Gbps devices?
2. Why does the STEC Mach16 100GB SLC suck so badly in comparison
to it's 200GB MLC cousin? I know that the 100GB drives won't
perform as well as the 200GB models but I did not expect this
much of a difference.
Even using bs=4096 on the Intel S3700, I was hoping to see 10K IOPS,
possibly matching the numbers from the anandtech review:
http://www.anandtech.com/show/6433/intel-ssd-dc-s3700-200gb-review/3
ATA-INTEL SSDSC2BA20-0270-186.31GB (MLC)
$ /usr/gnu/bin/dd if=/dev/zero of=file bs=4096 oflag=sync
$ iostat -cnx 1
r/sw/s kr/skw/s wait actv wsvc_t asvc_t %w %b
0.0 8714.50.0 34858.0 0.0 0.20.00.0 1 22
0.0 8410.90.0 33639.6 0.0 0.20.00.0 1 21
0.0 8431.10.0 33728.2 0.0 0.20.00.0 1 21
0.0 8295.00.0 33224.0 0.0 0.20.00.0 1 21
0.0 8970.10.0 35740.4 0.0 0.20.00.0 1 23
Am I getting the best possible performance out of the Intel S3700?
--
albert chin (ch...@thewrittenword.com)
___
openindiana-discuss mailing list
openindiana-discuss@openindiana.org
http://openindiana.org/mailman/listinfo/openindiana-discuss
___
openindiana-discuss mailing list
openindiana-discuss@openindiana.org
http://openindiana.org/mailman/listinfo/openindiana-discuss
