Raid levels

30 Jun 2009 7:15 AM

tshad

I am trying to get ready for my 70-431 test and am really confused on Raid
levels. I know what they are but I am using different books and tests and
they all say different things.

Raid 0 - fast but no fault tolerance.(stripe)
Raid 1 - not as fast but fault tolerance (Mirror)
Raid 5 - Better performance that 1 and better fault tolerance (Stripe and
parity)
Raid 10 - Faster than Raid 5 (Striping and Mirroring)

In the Microsoft book, it says that
Raid 10 best for log, Data and index files
If no Raid 10, then Store Data and Index file on Raid 5 with Tran logs on
Raid 10.

Exam Cram:
Raid 0 is good for Data
Raid 1 is good for Logs
Raid 5 should not be used in production ??????

One of the sample tests I am using has
Volume C - Raid 1 (2 x 10GB)
Volume D - Raid 1 (2 x 10GB)
Volume E - Raid 0. (2 x 20GB)

OS System on C:
Data on D:
Log on C:

How are you supposed to take a test when all your resources contradict each
other???? None of these resource agree on anything.

I have no idea what to use here.

I normally would use Mirrors for OS and Raid 5 for everything else. But I
have no idea what the test would expect.

Thanks,

Tom

30 Jun 2009 1:30 PM

Rick Sawtell

tshad, think about what you get with each of the RAID levels and then make
your decisions based on that information.

RAID 0 -- Striping only. This is the fastest disk access method, but
provides NO fault tolerance, and maximizes storage capacity.
RAID 1 -- Mirroring. This is not as fast as RAID 0 because a single
controller is writing to two disks. This does give fault tolerance, but uses
twice as much disk space.
RAID 2 -- Duplexing (Mirroring, but uses two separate disk controllers, so
is a bit faster than RAID 0 and adds some hardware fault tolerance as well)
RAID 3-4 ...
RAID 5 -- Striping with Parity. This provides fault tolerance, but the cost
is both speed (creating the parity checksums) as well as storage space. It
uses about 1 disks worth of space to store the parity checksums.
RAID 6 -- Same as 5 with a hot-swappable spare drive that automatically
activates.
RAID 1+0, 0+1, 10 -- This is the most expensive solution,but provides the
fastest access and highest level of fault tolerance. You essentially stripe
a set of disks and then mirror that stripe set.

Best practices suggest the following:

Put the OS on your local C drive.
You will want fast disks for your data retrieval and updating as well as
fault tolerance. If money is not an issue then RAID 10, otherwise, RAID 5.
Your transaction logs are sequential in nature, using a RAID 5 for
sequential access is not really what RAID 5 is best at. For that sequential
access methodology, you will want fast disks that are not performing parity
checksums and so forth. In that case, RAID 1 or RAID 10 is the best option
(fault tolerance, and good speed, and good sequential access).

When you study for the exam, takes all of these things in to account.

Speed, access style, money, fault tolerance and you should be able to derive
a good solution.

I hope this helps.

Rick Sawtell
MCT, MCSD, MCDBA, MCITPro: Database Admin/Impl

30 Jun 2009 7:49 PM

tshad

"Rick Sawtell" <r_sawtell@nospam.hotmail.com> wrote in message
news:ukEaxbY%23JHA.1376@TK2MSFTNGP02.phx.gbl...
> tshad, think about what you get with each of the RAID levels and then make
> your decisions based on that information.
>

I understand that.

But obviously there are other opinions and the opinion that I am concerned
with are what the test wants. And if ALL the study material out there
disagrees - there is NO WAY that I can see how to make the right choice.
Based on all you list out, you would think the study material would be
consistant but it isn't.

So really I am left with only flipping a coin.

Thanks,

Tom

Show quoteHide quote

> RAID 0 -- Striping only. This is the fastest disk access method, but
> provides NO fault tolerance, and maximizes storage capacity.
> RAID 1 -- Mirroring. This is not as fast as RAID 0 because a single
> controller is writing to two disks. This does give fault tolerance, but
> uses twice as much disk space.
> RAID 2 -- Duplexing (Mirroring, but uses two separate disk controllers, so
> is a bit faster than RAID 0 and adds some hardware fault tolerance as
> well)
> RAID 3-4 ...
> RAID 5 -- Striping with Parity. This provides fault tolerance, but the
> cost is both speed (creating the parity checksums) as well as storage
> space. It uses about 1 disks worth of space to store the parity checksums.
> RAID 6 -- Same as 5 with a hot-swappable spare drive that automatically
> activates.
> RAID 1+0, 0+1, 10 -- This is the most expensive solution,but provides the
> fastest access and highest level of fault tolerance. You essentially
> stripe a set of disks and then mirror that stripe set.
>
> Best practices suggest the following:
>
> Put the OS on your local C drive.
> You will want fast disks for your data retrieval and updating as well as
> fault tolerance. If money is not an issue then RAID 10, otherwise, RAID
> 5.
> Your transaction logs are sequential in nature, using a RAID 5 for
> sequential access is not really what RAID 5 is best at. For that
> sequential access methodology, you will want fast disks that are not
> performing parity checksums and so forth. In that case, RAID 1 or RAID 10
> is the best option (fault tolerance, and good speed, and good sequential
> access).
>
> When you study for the exam, takes all of these things in to account.
>
> Speed, access style, money, fault tolerance and you should be able to
> derive a good solution.
>
> I hope this helps.
>
> Rick Sawtell
> MCT, MCSD, MCDBA, MCITPro: Database Admin/Impl
>
>
>
>

30 Jun 2009 8:30 PM

Bob Simms

Show quote Hide quote

"tshad" <t***@pdsa.com> wrote in message
news:uPNhwvb#JHA.2872@TK2MSFTNGP05.phx.gbl...
>
> "Rick Sawtell" <r_sawtell@nospam.hotmail.com> wrote in message
> news:ukEaxbY%23JHA.1376@TK2MSFTNGP02.phx.gbl...
>> tshad, think about what you get with each of the RAID levels and then
>> make your decisions based on that information.
>>
> I understand that.
>
> But obviously there are other opinions and the opinion that I am concerned
> with are what the test wants. And if ALL the study material out there
> disagrees - there is NO WAY that I can see how to make the right choice.
> Based on all you list out, you would think the study material would be
> consistant but it isn't.
>
> So really I am left with only flipping a coin.
>

If you care about the data, never put it on RAID 0, For most of the other
questions, it will depend on the question in the exam. For example, if the
question says you need the fastest performance and maximum safety, raid 0+1.
If you want safety and you are limited in the disks you have, RAID 5. As
always, you have to provide the best answer given the scenario in the
particular question. There isn't a law as of the Meades and the Persians.
It varies depending on the scenario.

30 Jun 2009 8:55 PM

Andrew J. Kelly

I agree with Bob. You should never put data or logs on Raid 0 in production
and if the exam prep says you should I would throw it in the trash
immediately. Also the test should provide enough other information to make
the decision easier if you know the capabilities of each Raid.

--
Andrew J. Kelly SQL MVP
Solid Quality Mentors

Show quoteHide quote

"tshad" <t***@pdsa.com> wrote in message
news:uPNhwvb%23JHA.2872@TK2MSFTNGP05.phx.gbl...
>
> "Rick Sawtell" <r_sawtell@nospam.hotmail.com> wrote in message
> news:ukEaxbY%23JHA.1376@TK2MSFTNGP02.phx.gbl...
>> tshad, think about what you get with each of the RAID levels and then
>> make your decisions based on that information.
>>
> I understand that.
>
> But obviously there are other opinions and the opinion that I am concerned
> with are what the test wants. And if ALL the study material out there
> disagrees - there is NO WAY that I can see how to make the right choice.
> Based on all you list out, you would think the study material would be
> consistant but it isn't.
>
> So really I am left with only flipping a coin.
>
> Thanks,
>
> Tom
>
>> RAID 0 -- Striping only. This is the fastest disk access method, but
>> provides NO fault tolerance, and maximizes storage capacity.
>> RAID 1 -- Mirroring. This is not as fast as RAID 0 because a single
>> controller is writing to two disks. This does give fault tolerance, but
>> uses twice as much disk space.
>> RAID 2 -- Duplexing (Mirroring, but uses two separate disk controllers,
>> so is a bit faster than RAID 0 and adds some hardware fault tolerance as
>> well)
>> RAID 3-4 ...
>> RAID 5 -- Striping with Parity. This provides fault tolerance, but the
>> cost is both speed (creating the parity checksums) as well as storage
>> space. It uses about 1 disks worth of space to store the parity
>> checksums.
>> RAID 6 -- Same as 5 with a hot-swappable spare drive that automatically
>> activates.
>> RAID 1+0, 0+1, 10 -- This is the most expensive solution,but provides the
>> fastest access and highest level of fault tolerance. You essentially
>> stripe a set of disks and then mirror that stripe set.
>>
>> Best practices suggest the following:
>>
>> Put the OS on your local C drive.
>> You will want fast disks for your data retrieval and updating as well as
>> fault tolerance. If money is not an issue then RAID 10, otherwise, RAID
>> 5.
>> Your transaction logs are sequential in nature, using a RAID 5 for
>> sequential access is not really what RAID 5 is best at. For that
>> sequential access methodology, you will want fast disks that are not
>> performing parity checksums and so forth. In that case, RAID 1 or RAID
>> 10 is the best option (fault tolerance, and good speed, and good
>> sequential access).
>>
>> When you study for the exam, takes all of these things in to account.
>>
>> Speed, access style, money, fault tolerance and you should be able to
>> derive a good solution.
>>
>> I hope this helps.
>>
>> Rick Sawtell
>> MCT, MCSD, MCDBA, MCITPro: Database Admin/Impl
>>
>>
>>
>>
>
>