Message-Id: <199705301318.GAA205570@dns.ccit.arizona.edu> Date: Fri, 30 May 1997 09:07:48 -0400 From: David Barber <mailto:david@OHIOLINK.EDU> Subject: Re: RAID To: mailto:IMAGELIB@LISTSERV.ARIZONA.EDU
Think of two things when you are looking at RAID, its theoretical performance issues and how its implementation can affect performance. Here are some thoughts on both ...RAID does have implications for performance, even if that was not the primary reason behind its development. In the case of some RAID levels, RAID 5 for example, you loose some performance with RAID because you have to write both the data and the associated parity information. When reading as well, performance can also degrade with certain RAID levels because you need to check the parity.
There are also RAID configurations like 0+1 which can enhance performance. 0+1 mirrors and stripes information across multiple drives. With 0+1, when you need information you can read part of it from both the two sets of drives. This configuration I am told also provides better write performance. Of course with RAID 5 you get integrity of storage with a much smaller cost than 0+1. To do 0+1, you have to buy twice the amount of storage.
There are many performance issues related to RAID other than just its theoretical performance impact. Here are some of the issues which can also impact perfomance with RAID:
1. Drive selection -- You can RAID different kinds of drives. These can be larger slower drives, e.g. 9GB drives that operate at lower RPMs, or smaller drives with higher speeds, say the new 10,000 RPM 4GB drives.
2. Interfaces -- Drives can connect to the host computer through a number of types of interfaces, e.g. SCSI, Ultra SCSI, Fiber Channel, etc. Newer generations of interface, e.g. Ultra SCSI, will give better performance.
3. Cable connections with the interface. Without getting too technical, the basic point here is that you can configure a RAID solution in different ways that provide more or less pipes over which data can travel to the host computer.
4. The design of the host computer -- you can attach a RAID system to a host computer with a slow internal bus (part of the computer through which input to the processor flows). The bus could be swamped if too much data is being moved through the bus.
There are other aspects to consider as well.
As somebody who is just about to buy more than 1 TB of RAID, I can tell you that you need help from experts when buying RAID. I have gotten a lot of help from our computer specialists and the storage technicians of the vendor I choose when I picked my configuration.
You also need to look closely at the application you are using. There are many instances when you have different parts of the data, some of which are used much more frequently than others and are more performance sensitive. For example, in my application, a large collection of textual information, I have the text index on 0+1 RAID of 4 GB drives to get better performance and put the text files themselves on 9 GB drives using RAID 5. Thus I was able to reduce my overall cost and enhance performance.
I have probably gone on for long enough on the subject.
David Barber Director, New Service Development OhioLINK
---------- > From: A. Harry Williams <mailto:HARRY@VM.MARIST.EDU>
> To: mailto:IMAGELIB@LISTSERV.ARIZONA.EDU
> Subject: Re: RAID
> Date: Thursday, May 29, 1997 11:16 PM
>
> On Thu, 29 May 1997 07:13:32 -0700 Mike Betz said:
> >RAID technology is a method of ensuring data integrity, not access
speed. > Technically, its a method of measuring the level of data integrity,
> levels 0-5, though most people use it to mean RAID level 5.
>
> >RAID is the preferred method for servers in general. What RAID provides
> >for is the ability of the server to continue to function during a hard
> >disk failure without loss of data. It is similar to disk mirrorring.
> disk mirrorring is one level of RAID.
>
> >
> >Access speed is something different, and usually costs. Some softwares
use > >caching to increase speed of access.
> RAID does have some implications for reduced access speed, which is why
> they often have cacheing built-in.
>
> >
> >On Thu, 29 May 1997, Jennifer Brasher wrote:
> >
> >Mike Betz
> >mailto:betz@wln.com
> >
> >> "ello,.
> >> I'm after some good articles about RAID technology for image servers.
> >> Anyone know of any recently published? I need to have these to
convince > >> people around here that images need a special server, faster. I deduce
RAID > >> is the way to go? Any opinions? Anyone heard of anything better? ...
in the > >> pipeline? Jennifer QCA.
> >>
> >> VIEWS EXPRESSED ARE PERSONAL,NOT UNIVERSITY POLICY.
> >>
>
>***************************************************************************
*** > >>
> >> Jennifer Brasher `` ` ` ` EMAIL: mailto:J.Brasher@ins.gu.edu.au
> >> Art Librarian `` ` ` VOICE: +61 (07) 3875 3132
> >> Image Manager ` ` `FAX: +61 (07) 3875 3133
> >>
> >> SNAIL MAIL ` ` ` `
> >> Queensland College of Art Library ` ` ` ` *
> >> Griffith University ``````` *
> >>
> >> Clearview Terrace ``*` *
> >> Morningside Campus *
> >> PO Box 84 Morningside Brisbane Queensland 4170 Australia
> >>
>
>***************************************************************************
*** > >>
> >>
>
>***************************************************************************
*** > >>