RE: Scsi I/O speed

Thanks Chris!

        I will have to look at your site, what is the URL? Is there information to check on w/b w/t cache settings? Isn't one of them better than the other? Also, this is not a RAID system, just a couple of SCSI's hangin out and being beat on.
KK

-----Original Message-----
Spence
Sent: Thursday, July 12, 2001 1:30 PM
To: Multiple recipients of list ORACLE-L

Buzzwords are great.

There are a few good raid papers on my site that cover some of those he mentioned.

w/t is write through cache
w/b is write back cache

Write back cache is basically battery backed up memory on the raid controller to provide exceptially high transactional rates.

Stripe size is the size used when creating arrays, it detirmines how the scsi controller writes out to disks in stripes.

OS block size is the obvious.

Command tagging is when you allow the OS to send multiple scsi commands to the controller to execute parallel. This in theory reduces the need to awknowledge each request and execute it.

Disconnect/reconnect support is another feature that allows the disk to disconnect to from the system when handling a load of requests. This generally reduces the bus utiliization.

This is a very brief run down of the things mentioned.

HTH, "Walking on water and developing software from a specification are easy if both are frozen."

Christopher R. Spence OCP MCSE MCP A+ RAPTOR CNA Oracle DBA
Phone: (978) 322-5744
Fax: (707) 885-2275

Fuelspot
73 Princeton Street
North, Chelmsford 01863

-----Original Message-----
Sent: Thursday, July 12, 2001 12:21 PM
To: Multiple recipients of list ORACLE-L

Ross,

        I love computers and I love tinckering with them to make them run sweetly, but I don't have a friggen CLUE about what you just wrote!:) Point me to the mythical paper where all of this vast knowledge and buzzwords came from and I will dive in:) KK

-----Original Message-----
Sent: Thursday, July 12, 2001 10:56 AM
To: Multiple recipients of list ORACLE-L

also, consider turning OFF command tag queueing....check mobo drivers for i/o bus-related hw....check w/t vice w/b cache, look at stripe stride vice os block size.....

Ross

-----Original Message-----
Sent: Wednesday, July 11, 2001 1:41 PM
To: Multiple recipients of list ORACLE-L

Kevin Kostyszyn wrote:

> Hi all,
> I was measuring the i/o performance of my scsi drives and I
> have a

quick
> question that maybe someone could shed some light upon. Currently I
> am using Ultra 2/Wide scsi conrollers, this is supposed to have an I/O
> of 80mb/s. Well, when I perform the test all of the machines seem to
> be operation at halp of the max speed. One operates at about 20mb/s
> read and write and the others are even slower than that. Now on the
> first one, it

is
> the only HD on the controller, on the others there are two disks.
> Even on my Ultra/160 it seems to be maxing out at 40 read and write.
> Am I missing something? Am I reading this the wrong way?
> Help:(
>
> Sincerely,
> Kevin Kostyszyn
> DBA
> Dulcian, Inc
> www.dulcian.com
> kevin_at_dulcian.com

Kevin,

download SandraSoft's benchmarking tools - download.com is a good place to start. There is quite a difference between SCSI controller interface speeds and actual trasfer speeds between the OS and the physical hard drive. The Interface speed is a theoretical max, and is more important when configuring several drives on a single controller channel - e.g. RAID 0, 0+1, 5, etc. If you have 4 drives on a channel configured as a 4 drive RAID 0 volume, the controller channel SCSI interface speed could be the rate-limiting-factor. (e.g. 4 drives with an *average* transfer rate of 25 MB/sec = 100 MB/sec > 80 MB/sec).

As there is a cache on the hard drive (2-4 MB is customary) and could be a cache on the RAID contoller (128 MB - 4 GB?) the channel should be saturated during memory to memory transfers (after negotiation for the transfer has taken place) - short bursts which are then slowed by the subsequent access of the phyiscal media.

Typical sustained read/write speeds are on the order of 30 MB/sec on the latest and greatest 10,000 RPM drives. The fastest sustained read/write I've seen is here - is for the 15,000 RPM Seagate Cheetah - close to 50 MB/sec on the outer tracks
http://www.storagereview.com/articles/200105/20010510ST373405LW_1.html

Interface                speed (MB/sec)
SCSI
Ultra Wide     UW          40
Ultra2 Wide    U2W         80
Ultra160       U160       160
Ultra320       U320       320

IDE
UDMA-33        ATA-4       33
UDMA-66        ATA-5       66
Ultra ATA      ATA-6      100

Most likely, seek time will dominate transfer time unless you hike the operating system IO_size up from 64 KB.

this site looks like fun:
http://www.storagereview.com/cgi-bin/bench_compare.pl

remember - little 'b' is bits, big 'B' is Bytes. This is extremely important if you happen to look at NAS - using Gigabit Ethernet for shared storage.

Paul

--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Paul Drake
  INET: paled_at_home.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in the
message BODY, include a line containing: UNSUB ORACLE-L (or the name of
mailing list you want to be removed from).  You may also send the HELP
command for other information (like subscribing).
--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Mohan, Ross
  INET: MohanR_at_STARS-SMI.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in the
message BODY, include a line containing: UNSUB ORACLE-L (or the name of
mailing list you want to be removed from).  You may also send the HELP
command for other information (like subscribing).

--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Kevin Kostyszyn
  INET: kevin_at_dulcian.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in the
message BODY, include a line containing: UNSUB ORACLE-L (or the name of
mailing list you want to be removed from).  You may also send the HELP
command for other information (like subscribing).
--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Christopher Spence
  INET: cspence_at_FuelSpot.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).

-- 
Please see the official ORACLE-L FAQ: http://www.orafaq.com
-- 
Author: Kevin Kostyszyn
  INET: kevin_at_dulcian.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from).  You may
also send the HELP command for other information (like subscribing).