70-229 and SQL file placement

I'm reading Thomas Moore's book for 70-229. It isn't as full of mistakes as
most study guides I have read, but some of what he says, I am just not too
sure about.

I'd like to get some feedback (arguments) going on SQL Server and file
placement. Of course, all this discussion should be based on Microsoft's
perceptions and the exam, which, I would bet money, are way out of step with
state of the art RAID systems in the real world.

Moore says that tables that are mostly constant but read a lot should be on
RAID 5. I think they should be on RAID 1, mirroring. Mirroring allows two
threads to access the disk mechanics simultaneously. RAID 5, on the other
hand, is faster for writing because it can drop off a few bytes on one
drive, then while the mechanics are doing their part, the controller has
already dropped off bytes to the other drives in the stripe. Do you agree?

He says that the log files should be on RAID 1 for reliability because they
are written to often. Well, any RAID except 0 gives data protection in the
event of a disk failure, but RAID 1 requires 2 writes for each single block
of data being saved. His reasoning is that calculating parity on a RAID 5
takes time. I think that calculating parity on a quality hardware RAID
controller will always be faster (a few hundred processor cycles at most)
than any hard drive with moving parts could possibly be. What do you think?

Or more important... what do you think Microsoft thinks?

Dale Preston
MCSE, MCDBA, MCAD

Dale

Moore is right; raid 1 is faster than raid 5. raid 1 takes no
calculation, just simply read and write with cloning the signal to a
second harddrive. your suggestion that a mirror set could handle two
threads is not true; al requests are send to both drives. That's why they
call it mirroring.

Raid 5 always asks some calculation, no matter how much hardware you use
for building the raid.every write request takes some calculation.

If you want good performance on your logs, use raid 0+1; four drives
forming two pairs of raid 0, mirroring each other.

my 2p.

jwk

"Dale" <> wrote in
news::

> I'm reading Thomas Moore's book for 70-229. It isn't as full of
> mistakes as most study guides I have read, but some of what he says, I
> am just not too sure about.
>
> I'd like to get some feedback (arguments) going on SQL Server and file
> placement. Of course, all this discussion should be based on
> Microsoft's perceptions and the exam, which, I would bet money, are
> way out of step with state of the art RAID systems in the real world.
>
> Moore says that tables that are mostly constant but read a lot should
> be on RAID 5. I think they should be on RAID 1, mirroring. Mirroring
> allows two threads to access the disk mechanics simultaneously. RAID
> 5, on the other hand, is faster for writing because it can drop off a
> few bytes on one drive, then while the mechanics are doing their part,
> the controller has already dropped off bytes to the other drives in
> the stripe. Do you agree?
>
> He says that the log files should be on RAID 1 for reliability because
> they are written to often. Well, any RAID except 0 gives data
> protection in the event of a disk failure, but RAID 1 requires 2
> writes for each single block of data being saved. His reasoning is
> that calculating parity on a RAID 5 takes time. I think that
> calculating parity on a quality hardware RAID controller will always
> be faster (a few hundred processor cycles at most) than any hard drive
> with moving parts could possibly be. What do you think?
>
> Or more important... what do you think Microsoft thinks?
>
> Dale Preston
> MCSE, MCDBA, MCAD
>
>
>

jwk

babbling on and on again "=?Utf-8?B?Qm9iYnk=?="
<> spewed in
news6619D3F-D816-4556-AF32-:

> Moore is right in saying that RAID 1 has better write performance but
> RAID 5 has excellent read performance.

but the point of an OLTP database would be to write almost as much as you
read. If you want decent performance in the real world I am still
sticking with my RAID 0+1 and if I am forced into using a raid 5 system
it better be for DSS or some crappy little db that no one will notice how
bad it performs for I,U,D...hardware or software RAID be damned, RAID 0+1
HW still kicks HW RAID 5's a$$...(and I would like to see you do SW RAID
0+1 on an MS box <VBEG>)

--
Neil MCNGP #30
Minutus cantorum, minutus balorum, minutus carborata descendum pantorum.

When I die, I want to die like my grandfather-who died peacefully in
his sleep. Not screaming like all the passengers in his car.

Neil

>but the point of an OLTP database would be to write almost as much as you
>read. If you want decent performance in the real world I am still
>sticking with my RAID 0+1 and if I am forced into using a raid 5 system
>it better be for DSS or some crappy little db that no one will notice how
>bad it performs for I,U,D...hardware or software RAID be damned, RAID 0+1
>HW still kicks HW RAID 5's a$$...(and I would like to see you do SW RAID
>0+1 on an MS box <VBEG>)

nah, tera ramsan the way to go!

Kline Sphere (Chalk) MCNGP #3

The Poster Formerly Known as Kline Sphere

babbling on and on again The Poster Formerly Known as Kline Sphere <.>
spewed in news::

> tera ramsan

ok, fine. then with conventional disks 0+1 kicks a$$.

--
Neil MCNGP #30
Minutus cantorum, minutus balorum, minutus carborata descendum pantorum.

When I die, I want to die like my grandfather-who died peacefully in
his sleep. Not screaming like all the passengers in his car.

Neil

>> tera ramsan
>
>ok, fine. then with conventional disks 0+1 kicks a$$.

touchy, touchy...

Kline Sphere (Chalk) MCNGP #3

The Poster Formerly Known as Kline Sphere

"Bobby" <> wrote in message
newsEDFA4D8-20D8-41C7-B94E-...

<snip repost of post that was a post similar to a post made 13 hours ago>

We get the point, Bobby. Good advice, just too much of it.

--

KB - MCNGP "silent thug" #26

first initial last name AT hotmail DOT com

Ken Briscoe