(no subject)

Oliver Lehmann lehmann at ans-netz.de
Wed Sep 30 06:28:45 UTC 2009

I got 4 new SATA disks (WD Green, 1TB, WD10EADS) I want to use to replace
my old 250GB disks attached to my 3ware controller.
I want to reuse the old 250GB disks in some systems running old PATA disks
ight now as system drives. So what I did now was gathering SATA performance
tatistics with the new 1TB drive to just find out what would be the maximum
performance I would get out of these disks to compare them later with my
3ware when they are configured as RAID-5. 

A colleague of mine has the same disks in a new Nvidia Atom 330 system and
he told me that he reaches around 70MB/sec write speed with a single large
file on a single disk running linux 2.6. 

I hooked the disk up to my client: 

FreeBSD 7.2-STABLE #0: Tue Jul 28 12:59:47 CEST 2009
CPU: AMD Athlon(tm) 64 Processor 3500+ (2200.10-MHz K8-class CPU)
usable memory = 2138615808 (2039 MB)
atapci0: <VIA 6420 SATA150 controller> port 
 -0xb4ff irq 20 at device 15.0 on pci0
ad4: 953869MB <WDC WD10EADS-00L5B1 01.01A01> at ata2-master SATA150 

because the on-board controller is a VIA 6420 I had to set the SATA150
Jumper on the harddisk to have the controller detect the drive. 

On FreeBSD I used gpart+gpt to create a 1TB partition and then simply ran
newfs /dev/ad4p1 and mounted the new filesystem afterwards.
I then ran a dd in=/dev/zero of=/mnt/tmp/test.dd bs=1M count=4069 and dd
reported me a write speed of around 25MB/sec. This made me feel kinda bad
so I gave bonnie++ a try. The result was: 

Version  1.96       ------Sequential Output------ --Sequential Input- 
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec 
kartoffel.salats 4G   548  96 28924   3 14617   2  1141  96 36869   3 199.7 
Latency               167ms   71702us    1759ms   23957us   75351us   2286ms
Version  1.96       ------Sequential Create------ --------Random 
kartoffel.salatschu -Create-- --Read--- -Delete-- -Create-- --Read--- 
             files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
                16   552   1 +++++ +++  1486   2   531   1 +++++ +++  1278 1
Latency             91403us     156us   28424us   22901us      87us  22820us

This also did not look that good comparing to the bonnie output the
colleague gave me from his shiny new ION system. 

I then booted the latest knoppix (Also a 2.6.whatever linux kernel),
created a filesystem on /dev/sd1a (mkfs.ext3 /dev/sd1a) and mounted the
filesystem as well. The same dd I ran on FreeBSD I also ran on Knoppix
and this gave me 57.3MB/sec (wow compared to 25MB/sec). I then also
started bonnie++ just to see that this one is also much better: 

Version  1.96       ------Sequential Output------ --Sequential Input- 
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec 
Microknoppix     4G   305  99 55905  18 31896   9   959  98 80414  10 211.7 
Latency             28579us    1075ms    1046ms   26376us   20962us    272ms
Version  1.96       ------Sequential Create------ --------Random 
Microknoppix        -Create-- --Read--- -Delete-- -Create-- --Read--- 
             files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
                16 27135  59 +++++ +++ +++++ +++ 29369  62 +++++ +++ +++++ 
Latency             23535us    9969us    9927us   11680us    1182us   9985us

Does anyone know if there is something I can tune on FreeBSD to get more
speed? hw.ata.wc is enabled of course. 

hw.ata.wc: 1
hw.ata.atapi_dma: 1
hw.ata.ata_dma_check_80pin: 1
hw.ata.ata_dma: 1 

I'll retest both setups with a plugged in Promise SATA300 PCI controller
but I doubt that it will get faster. I tried the controller before, and
on an dual PIII-850 system with L440GX chipset and 2GB of RAM the
controller gave me around 40MB/sec on write and on my amd64 system I also
only got around 25MB/sec (even this makes no sense to me why my old PIII
is faster then my much newer amd64) but I'll come back with better
numbers for this controller later.

More information about the freebsd-performance mailing list