implement Raid 10 as a span (concatenation) of mirrors not a stripe.
I have 22 disks and over half are idle.
I'm having to split the database file and log file into pieces.
Paul
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:ucSRJy05DHA.2560@.TK2MSFTNGP09.phx.gbl...
quote:
> But she can't do a RAID 10 with just 3 drives. It's either a Raid 1 and
you
quote:Paul,
> loose a drive or Raid 5.
> --
> Andrew J. Kelly
> SQL Server MVP
>
> <sightdev@.yahoo.com> wrote in message
> news:9861f5eb.0401300736.6761b95d@.posting.google.com...
>
Not sure what you mean exactly. The RAID numbers often get confusing when
it comes to 1+0 or 0+1. One will be a stripe of mirrors and the other is a
mirror of stripes. The difference is mainly in the redundancy as the stripe
of mirrors will tolerate potentially more failures than a mirror of stripes.
In either case all the disks are used but you only get half the usable disk
space. That is how Raid arrays work. You sacrifice drives for redundancy
and speed. I don't know why you say you need to split your files. That
should not be necessary.
Andrew J. Kelly
SQL Server MVP
"Paul Cahill" <XYZpaul.cahillXYZ@.blueyonder.co.uk> wrote in message
news:eyHa3nY6DHA.2264@.tk2msftngp13.phx.gbl...
quote:|||You are correct on the raid levels and raid 10 is defined as a stripe of
> A word of warning. I discovered my expensive Dell Perc3/DC controllers
> implement Raid 10 as a span (concatenation) of mirrors not a stripe.
> I have 22 disks and over half are idle.
> I'm having to split the database file and log file into pieces.
> Paul
>
> "Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
> news:ucSRJy05DHA.2560@.TK2MSFTNGP09.phx.gbl...
> you
>
mirrors. However Dell's documentation for the Perc3 and other controllers
now states (and it's true) that their implementation of raid 10 is a span of
mirrors. ie if you have a raid 10 of four 18GB disks and you create a 9GB
log file the file will wholy reside on the first mirror pair. The second
mirror pair will not be used until the 1st mirror pair is full.
This explains the dissappointing performance we have from our 22 disk
system. Although I benchmarked it, the one thing I did not do when I set it
up was to watch the disk lights.
For example our log file does indeed live on a 4 disk raid 10 array of 18GB
15K drives. The log file is smaller than 18GB. When I/O occurs only two of
the disk lights flash. The other two never flash. If it was true raid 10
then all 4 lights should flash as the i/o is striped.
So I'm having to do the striping myself by spitting the files across raid
arrays. ie Indexes on one, certain tables on another. Not a bad thing
anyway I guess. But if I'd known about dell's Raid 10 I'd have planned
differently. It's not east to make changes on a 24/7 system.
An alternative with the dell perc3 would be to stripe the drives (Raid 0)
and then mirror using the O/S. Not one I care to choose.
Paul
http://docs.us.dell.com/docs/storag...g/en/Levels.htm
The above document describes Raid 10 on the perc3/dc as a stripe of mirrors
http://docs.us.dell.com/docs/softwa...32/ch8_perc.htm
This document shows that perc3/dc implements as a concatenation
Search for string "Supported RAID Levels for PERC Subsystem Controllers"
http://docs.us.dell.com/docs/softwa...tor.htm#1037893
This document describes spanned disks
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:uXdx2uZ6DHA.2952@.TK2MSFTNGP09.phx.gbl...
quote:
> Paul,
> Not sure what you mean exactly. The RAID numbers often get confusing when
> it comes to 1+0 or 0+1. One will be a stripe of mirrors and the other is
a
quote:
> mirror of stripes. The difference is mainly in the redundancy as the
stripe
quote:
> of mirrors will tolerate potentially more failures than a mirror of
stripes.
quote:
> In either case all the disks are used but you only get half the usable
disk
quote:|||Paul,
> space. That is how Raid arrays work. You sacrifice drives for redundancy
> and speed. I don't know why you say you need to split your files. That
> should not be necessary.
> --
> Andrew J. Kelly
> SQL Server MVP
>
> "Paul Cahill" <XYZpaul.cahillXYZ@.blueyonder.co.uk> wrote in message
> news:eyHa3nY6DHA.2264@.tk2msftngp13.phx.gbl...
and[QUOTE]
>
I have to admit I don't know exactly what they are referring to. Hopefully
I can read more on it later but it doesn't sound pretty. I would call and
talk to someone in tech support directly who says claims they know what they
are talking about when it comes to this implementation of RAID 10 to get his
take. If it's anything close to what the docs state I would ask for my
money back since that is no where close to the definition of RAID 10 that I
have ever seen and I don't know how they can even claim it as such. From
what I gather the reason why the other disks lights are not blinking is that
with "concatenation" it fills up the first disk before it moves to the next
and you probably don't have enough data on it yet. Hopefully we are just
misreading their information but in either case I wish you luck.
Andrew J. Kelly
SQL Server MVP
"Paul Cahill" <XYZpaul.cahillXYZ@.blueyonder.co.uk> wrote in message
news:OYmMKUj6DHA.1936@.TK2MSFTNGP12.phx.gbl...
quote:
> You are correct on the raid levels and raid 10 is defined as a stripe of
> mirrors. However Dell's documentation for the Perc3 and other controllers
> now states (and it's true) that their implementation of raid 10 is a span
of
quote:
> mirrors. ie if you have a raid 10 of four 18GB disks and you create a 9GB
> log file the file will wholy reside on the first mirror pair. The second
> mirror pair will not be used until the 1st mirror pair is full.
> This explains the dissappointing performance we have from our 22 disk
> system. Although I benchmarked it, the one thing I did not do when I set
it
quote:
> up was to watch the disk lights.
> For example our log file does indeed live on a 4 disk raid 10 array of
18GB
quote:
> 15K drives. The log file is smaller than 18GB. When I/O occurs only two of
> the disk lights flash. The other two never flash. If it was true raid 10
> then all 4 lights should flash as the i/o is striped.
> So I'm having to do the striping myself by spitting the files across raid
> arrays. ie Indexes on one, certain tables on another. Not a bad thing
> anyway I guess. But if I'd known about dell's Raid 10 I'd have planned
> differently. It's not east to make changes on a 24/7 system.
> An alternative with the dell perc3 would be to stripe the drives (Raid 0)
> and then mirror using the O/S. Not one I care to choose.
> Paul
> http://docs.us.dell.com/docs/storag...g/en/Levels.htm
> The above document describes Raid 10 on the perc3/dc as a stripe of
mirrors
quote:|||I am taking this up with Dell, others are discussing this in the Dell Talk
>
> http://docs.us.dell.com/docs/softwa...32/ch8_perc.htm
> This document shows that perc3/dc implements as a concatenation
> Search for string "Supported RAID Levels for PERC Subsystem Controllers"
>
> http://docs.us.dell.com/docs/softwa...tor.htm#1037893
> This document describes spanned disks
>
> "Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
> news:uXdx2uZ6DHA.2952@.TK2MSFTNGP09.phx.gbl...
when[QUOTE]
is[QUOTE]
> a
> stripe
> stripes.
> disk
redundancy[QUOTE]
> and
makes[QUOTE]
>
forum.
The sales guys and the 1st line tech support people don't even understand
the issue.
At the moment I'm getting unfufilled call back promises.
My main data file is on a (Dell) raid 10 of 6 disks 18GB disks. The main
data file is over 36GB. If I watch the lights on this logical array they do
all flash but not together. You can see it happening in pairs. If I/O on the
data file was truly random then I would be getting some bangs for all my
bucks but of course there are always hotspots in data access.
No single i/o will ever proceed faster than the speed of a single 18GB 15K
drive.
The only way I can see around this is to stripe using the perc3 and then
mirror the stripe using the operating system. Not a route I am prepared to
proceed with.
Paul
"Andrew J. Kelly" <sqlmvpnooospam@.shadhawk.com> wrote in message
news:eImaq%23l6DHA.2480@.TK2MSFTNGP10.phx.gbl...
quote:
> Paul,
> I have to admit I don't know exactly what they are referring to.
Hopefully
quote:
> I can read more on it later but it doesn't sound pretty. I would call and
> talk to someone in tech support directly who says claims they know what
they
quote:
> are talking about when it comes to this implementation of RAID 10 to get
his
quote:
> take. If it's anything close to what the docs state I would ask for my
> money back since that is no where close to the definition of RAID 10 that
I
quote:
> have ever seen and I don't know how they can even claim it as such. From
> what I gather the reason why the other disks lights are not blinking is
that
quote:
> with "concatenation" it fills up the first disk before it moves to the
next
quote:
> and you probably don't have enough data on it yet. Hopefully we are just
> misreading their information but in either case I wish you luck.
>
>
> --
> Andrew J. Kelly
> SQL Server MVP
>
> "Paul Cahill" <XYZpaul.cahillXYZ@.blueyonder.co.uk> wrote in message
> news:OYmMKUj6DHA.1936@.TK2MSFTNGP12.phx.gbl...
controllers[QUOTE]
span[QUOTE]
> of
9GB[QUOTE]
second[QUOTE]
> it
> 18GB
of[QUOTE]
raid[QUOTE]
0)[QUOTE]
> mirrors
http://docs.us.dell.com/docs/softwa...tor.htm#1037893[QUOTE]
> when
> is
> redundancy
That[QUOTE]
controllers[QUOTE]
1[QUOTE]
> makes
>
No comments:
Post a Comment