Disk Performance - Then & Now..

classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

Disk Performance - Then & Now..

David C. Rankin
Listmates,

        I ran a comparison on an old file server just to check disk performance of the
backup machine. I compared it to a new machine and the difference validates
Mohr's law:

[AMD K62-450 w/2 20G IBM Deskstar 7200 rpm drives]

sudo hdparm -tT /dev/sda
/dev/sda:
         Timing cached reads:    72 MB in  2.00 seconds =  35.94 MB/sec
         Timing buffered disk reads:   26 MB in  3.09 seconds =   8.41 MB/sec


[AMD Phenom 9850 w/2 500G Seagates in RAID1]

/dev/sda:
         Timing cached reads:   3504 MB in  2.00 seconds = 1751.75 MB/sec
         Timing buffered disk reads:  188 MB in  3.02 seconds =  62.33 MB/sec
/dev/sdb:
         Timing cached reads:   3494 MB in  2.00 seconds = 1746.89 MB/sec
         Timing buffered disk reads:  200 MB in  3.01 seconds =  66.53 MB/sec

/dev/dm-0:
         Timing cached reads:   3432 MB in  2.00 seconds = 1716.32 MB/sec
         Timing buffered disk reads:  194 MB in  3.02 seconds =  64.27 MB/sec

The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.

        The results are astounding when you think what has happened in just a few years...

--
David C. Rankin, J.D., P.E.
Rankin Law Firm, PLLC
510 Ochiltree Street
Nacogdoches, Texas 75961
Telephone: (936) 715-9333
Facsimile: (936) 715-9339
www.rankinlawfirm.com
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

Sandy Drobic
David C. Rankin wrote:

> Listmates,
>
> I ran a comparison on an old file server just to check disk performance of the
> backup machine. I compared it to a new machine and the difference validates
> Mohr's law:
>
> [AMD K62-450 w/2 20G IBM Deskstar 7200 rpm drives]
>
> sudo hdparm -tT /dev/sda
> /dev/sda:
> Timing cached reads:    72 MB in  2.00 seconds =  35.94 MB/sec
> Timing buffered disk reads:   26 MB in  3.09 seconds =   8.41 MB/sec
>
>
> [AMD Phenom 9850 w/2 500G Seagates in RAID1]
>
> /dev/sda:
> Timing cached reads:   3504 MB in  2.00 seconds = 1751.75 MB/sec
> Timing buffered disk reads:  188 MB in  3.02 seconds =  62.33 MB/sec
> /dev/sdb:
> Timing cached reads:   3494 MB in  2.00 seconds = 1746.89 MB/sec
> Timing buffered disk reads:  200 MB in  3.01 seconds =  66.53 MB/sec
>
> /dev/dm-0:
> Timing cached reads:   3432 MB in  2.00 seconds = 1716.32 MB/sec
> Timing buffered disk reads:  194 MB in  3.02 seconds =  64.27 MB/sec
>
> The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.

That is probably due to getting rid of the pci bus. Here's a hardware raid
(3ware 9650SE, 5 x 750 GB Seagate SATA2):

mordor:~ # hdparm -tT /dev/mapper/system-var

/dev/mapper/system-var:
 Timing cached reads:   13236 MB in  2.00 seconds = 6626.75 MB/sec
 Timing buffered disk reads:  1130 MB in  3.00 seconds = 376.05 MB/sec



--
Sandy

List replies only please!
Please address PMs to: news-reply2 (@) japantest (.) homelinux (.) com
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

Sandy Drobic
Sandy Drobic wrote:
> David C. Rankin wrote:

>> /dev/dm-0:
>> Timing cached reads:   3432 MB in  2.00 seconds = 1716.32 MB/sec
>> Timing buffered disk reads:  194 MB in  3.02 seconds =  64.27 MB/sec
>>
>> The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.
>
> That is probably due to getting rid of the pci bus. Here's a hardware raid
> (3ware 9650SE, 5 x 750 GB Seagate SATA2):
>
> mordor:~ # hdparm -tT /dev/mapper/system-var
>
> /dev/mapper/system-var:
>  Timing cached reads:   13236 MB in  2.00 seconds = 6626.75 MB/sec
>  Timing buffered disk reads:  1130 MB in  3.00 seconds = 376.05 MB/sec
>

Here's the data of my old server (old pci hardware raid, 3 x 18 GB SCSI):

/dev/rd/c0d0p3:
 Timing cached reads:   298 MB in  2.00 seconds = 148.70 MB/sec
 Timing buffered disk reads:   62 MB in  3.02 seconds =  20.56 MB/sec

Quite a difference. (^-^)

--
Sandy

List replies only please!
Please address PMs to: news-reply2 (@) japantest (.) homelinux (.) com
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

Per Jessen-2
Sandy Drobic wrote:

>> That is probably due to getting rid of the pci bus. Here's a hardware
>> raid (3ware 9650SE, 5 x 750 GB Seagate SATA2):
>>
>> mordor:~ # hdparm -tT /dev/mapper/system-var
>>
>> /dev/mapper/system-var:
>>  Timing cached reads:   13236 MB in  2.00 seconds = 6626.75 MB/sec
>>  Timing buffered disk reads:  1130 MB in  3.00 seconds = 376.05
>>  MB/sec
>>
>
> Here's the data of my old server (old pci hardware raid, 3 x 18 GB
> SCSI):
>
> /dev/rd/c0d0p3:
>  Timing cached reads:   298 MB in  2.00 seconds = 148.70 MB/sec
>  Timing buffered disk reads:   62 MB in  3.02 seconds =  20.56 MB/sec
>
> Quite a difference. (^-^)

The SCSI transfer speed also matters a lot - those 18Gb drives were
pretty slow when compared to todays U320 drives.



/Per Jessen, Z├╝rich

--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

Felix Miata
In reply to this post by David C. Rankin
On 2008/10/23 02:48 (GMT-0500) David C. Rankin composed:

> I ran a comparison on an old file server just to check disk performance of the
> backup machine. I compared it to a new machine and the difference validates
> Mohr's law:

> [AMD K62-450 w/2 20G IBM Deskstar 7200 rpm drives]

> sudo hdparm -tT /dev/sda
> /dev/sda:
> Timing cached reads:    72 MB in  2.00 seconds =  35.94 MB/sec
> Timing buffered disk reads:   26 MB in  3.09 seconds =   8.41 MB/sec

Almost certainly UDMA33, which dates back to around 11 years ago. I'm not
aware of any Socket 7 chipsets that supported UDMA66. If your K6/2-450 was
state of the art speed when its motherboard was acquired, it would probably
have pre-dated UDMA66 anyway.

> [AMD Phenom 9850 w/2 500G Seagates in RAID1]

> /dev/sda:
> Timing cached reads:   3504 MB in  2.00 seconds = 1751.75 MB/sec
> Timing buffered disk reads:  188 MB in  3.02 seconds =  62.33 MB/sec
> /dev/sdb:
> Timing cached reads:   3494 MB in  2.00 seconds = 1746.89 MB/sec
> Timing buffered disk reads:  200 MB in  3.01 seconds =  66.53 MB/sec

> /dev/dm-0:
> Timing cached reads:   3432 MB in  2.00 seconds = 1716.32 MB/sec
> Timing buffered disk reads:  194 MB in  3.02 seconds =  64.27 MB/sec

SATA 300 probably, which itself needs a test of Moore's law. I doubt you'll
find a doubling over real world performance of an equivalent system only
18-24 months older.

> The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.

> The results are astounding when you think what has happened in just a few years...

More like 9 years (UDMA33 to SATA300), in which of late the law has been
breaking down.

Still, what we have now begs the question how fast is fast enough. I stopped
being impressed with "faster" 5 years ago, around the time gigabit ethernet
and then DDR2 RAM hit the street. Even with FIOS the internet is most often
the bottleneck noticed by the face in front of the display, and a big
bottleneck it is.
--
"Everyone should be quick to listen, slow to speak, and
slow to become angry." James 1:19 NIV

 Team OS/2 ** Reg. Linux User #211409

Felix Miata  ***  http://fm.no-ip.com/
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

Larry Stotler
On Thu, Oct 23, 2008 at 8:15 AM, Felix Miata <[hidden email]> wrote:
> On 2008/10/23 02:48 (GMT-0500) David C. Rankin composed:
>> The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.
>
>>       The results are astounding when you think what has happened in just a few years...
>
> More like 9 years (UDMA33 to SATA300), in which of late the law has been
> breaking down.

Here's my results from an old HP NetServer E60  Dual P3/500Mhz, 256MB RAM:

Stand alone SCSI drive:

/dev/sda:
       Timing cached reads:   258 MB in  2.00 seconds = 129.00 MB/sec
       Timing buffered disk reads:  92 MB in  3.04 seconds =  30.26 MB/sec

Software Raid 5(on disks 2-4)

/dev/sdb:
       Timing cached reads:   258 MB in  2.00 seconds = 129.00 MB/sec
       Timing buffered disk reads:  90 MB in  3.02 seconds =  29.84 MB/sec

As you can see, the software RAID 5 was slightly slower than accessing
the standalone disk.  The drives are basically limited by the fact
that I am currently using the onboard SCSI controller, which is only
40MB/s.  sda and sdb are Fujitsu MAJ3091MP(10k rpm, U160, 4k buffer)
and sdc and sdd are HP A 68-F305(Which I think are similar specs).

However, since this machine is a data backup server, speed isn't
really important because everything sent to it comes in over 100Mb
ethernet right now.  If I move it to gigabit, then I may move it to a
faster PCI SCSI card.....
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

David C. Rankin
In reply to this post by Sandy Drobic
Sandy Drobic wrote:

> Sandy Drobic wrote:
>> David C. Rankin wrote:
>
>>> /dev/dm-0:
>>> Timing cached reads:   3432 MB in  2.00 seconds = 1716.32 MB/sec
>>> Timing buffered disk reads:  194 MB in  3.02 seconds =  64.27 MB/sec
>>>
>>> The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.
>> That is probably due to getting rid of the pci bus. Here's a hardware raid
>> (3ware 9650SE, 5 x 750 GB Seagate SATA2):
>>
>> mordor:~ # hdparm -tT /dev/mapper/system-var
>>
>> /dev/mapper/system-var:
>>  Timing cached reads:   13236 MB in  2.00 seconds = 6626.75 MB/sec
>>  Timing buffered disk reads:  1130 MB in  3.00 seconds = 376.05 MB/sec
>>
>
> Here's the data of my old server (old pci hardware raid, 3 x 18 GB SCSI):
>
> /dev/rd/c0d0p3:
>  Timing cached reads:   298 MB in  2.00 seconds = 148.70 MB/sec
>  Timing buffered disk reads:   62 MB in  3.02 seconds =  20.56 MB/sec
>
> Quite a difference. (^-^)
>

Sandy,

        Are those 5 750G drives are in RAID0/RAID5 ?  That is incredible disk performance.

--
David C. Rankin, J.D., P.E.
Rankin Law Firm, PLLC
510 Ochiltree Street
Nacogdoches, Texas 75961
Telephone: (936) 715-9333
Facsimile: (936) 715-9339
www.rankinlawfirm.com
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Disk Performance - Then & Now..

Sandy Drobic
David C. Rankin wrote:

> Sandy Drobic wrote:
>> Sandy Drobic wrote:
>>> David C. Rankin wrote:
>>>> /dev/dm-0:
>>>> Timing cached reads:   3432 MB in  2.00 seconds = 1716.32 MB/sec
>>>> Timing buffered disk reads:  194 MB in  3.02 seconds =  64.27 MB/sec
>>>>
>>>> The Software RAID penalty -- virtually non-existent as far as hdparm is concerned.
>>> That is probably due to getting rid of the pci bus. Here's a hardware raid
>>> (3ware 9650SE, 5 x 750 GB Seagate SATA2):
>>>
>>> mordor:~ # hdparm -tT /dev/mapper/system-var
>>>
>>> /dev/mapper/system-var:
>>>  Timing cached reads:   13236 MB in  2.00 seconds = 6626.75 MB/sec
>>>  Timing buffered disk reads:  1130 MB in  3.00 seconds = 376.05 MB/sec
>>>
>> Here's the data of my old server (old pci hardware raid, 3 x 18 GB SCSI):
>>
>> /dev/rd/c0d0p3:
>>  Timing cached reads:   298 MB in  2.00 seconds = 148.70 MB/sec
>>  Timing buffered disk reads:   62 MB in  3.02 seconds =  20.56 MB/sec
>>
>> Quite a difference. (^-^)
>>
> Are those 5 750G drives are in RAID0/RAID5 ?  That is incredible disk performance.

Yes, it's a RAID5. Though write performance is much less. Most of the time
this is theoretical data with no real life worth. In most cases it is more
important how fast many small files are written or read.

Still, since the controller is working in write-back mode, write performance
is acceptable in my case.




--
Sandy

List replies only please!
Please address PMs to: news-reply2 (@) japantest (.) homelinux (.) com
--
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]