PERC 4 performance

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

We just purchased a new 1850 server (2x3.4 Ghz) with 8GB of memory and
a dual channel PERC 4e controller (internal drives are on their own
controller); each channel is connected to 6 36GB 15K drives for a total
of 12. I had our server folks build 3 RAID 0/1 arrays of 4 disks each.
We're running Windows 2000 Advanced, and this is going to be a database
server (running MS SQL Server 2k).

While setting up my databases, I had occasion to copy a 21 GB file from
one array to another. Using standard Windows drag and drop, I'm
averaging only 30-33MB per second for this copy with no other load on
the disks or system. This seems kind of slow to me based on typical
hard disk transfer rates I've seen on storagereview.com and similar
sites; I'd expect at least twice this amount.

Are my expectations unreasonable? Does Windows have that much of an
overhead? Or is there possibly an issue here?
Reply here on the newsgroup please.

Thanks,

Steve

 

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

No, not RAID 1, RAID 0/1. RAID 5 is not necessarily a good thing on a
DB server since write performance usually suffers.

Steve

 

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

Thanks, but I'm quite familiar with different RAID levels. And that
site neglects to point out that RAID 5 arrays typically take a
performance hit due to the need to calculate parity for every block
that's written. While calculating parity doesn't very hard, AFAIK there
are no controllers extant that can write a RAID 5 array as fast as a
RAID 0/1 (or RAID 10, if you prefer that nomenclature) array. In the
database world we generally try to avoid RAID 5 if we have enough disks
(although many people use them on DSS (decision support [reporting])
systems where write performance is secondary to maximizing storage
space).

None of which, of course, answers my original question of whether my
arrays are functioning slower than they should be.

Steve

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

you probably want raid 5 for a database application... faster than raid 1
and more efficient in terms of storage space.
http://www.pcguide.com/ref/hdd/perf/raid/levels/single.htm

"Steve" <swechsler@bigfoot.com> wrote in message
news:1102699770.017922.20760@c13g2000cwb.googlegroups.com...
> We just purchased a new 1850 server (2x3.4 Ghz) with 8GB of memory and
> a dual channel PERC 4e controller (internal drives are on their own
> controller); each channel is connected to 6 36GB 15K drives for a total
> of 12. I had our server folks build 3 RAID 0/1 arrays of 4 disks each.
> We're running Windows 2000 Advanced, and this is going to be a database
> server (running MS SQL Server 2k).
>
> While setting up my databases, I had occasion to copy a 21 GB file from
> one array to another. Using standard Windows drag and drop, I'm
> averaging only 30-33MB per second for this copy with no other load on
> the disks or system. This seems kind of slow to me based on typical
> hard disk transfer rates I've seen on storagereview.com and similar
> sites; I'd expect at least twice this amount.
>
> Are my expectations unreasonable? Does Windows have that much of an
> overhead? Or is there possibly an issue here?
> Reply here on the newsgroup please.
>
> Thanks,
>
> Steve
>

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

Like hell he does. 0+1 offers the highest performance, and redundancy. its
also the most expsensive. But its a very HOT setup for a DB server, where
Data I/O is a huge performance factor.

- NuTs

"Christopher Muto" <muto@worldnet.att.net> wrote in message
news:6olud.5133$t44.3693@trndny03...
> you probably want raid 5 for a database application... faster than raid 1
> and more efficient in terms of storage space.
> http://www.pcguide.com/ref/hdd/perf/raid/levels/single.htm
>
> "Steve" <swechsler@bigfoot.com> wrote in message
> news:1102699770.017922.20760@c13g2000cwb.googlegroups.com...
> > We just purchased a new 1850 server (2x3.4 Ghz) with 8GB of memory and
> > a dual channel PERC 4e controller (internal drives are on their own
> > controller); each channel is connected to 6 36GB 15K drives for a total
> > of 12. I had our server folks build 3 RAID 0/1 arrays of 4 disks each.
> > We're running Windows 2000 Advanced, and this is going to be a database
> > server (running MS SQL Server 2k).
> >
> > While setting up my databases, I had occasion to copy a 21 GB file from
> > one array to another. Using standard Windows drag and drop, I'm
> > averaging only 30-33MB per second for this copy with no other load on
> > the disks or system. This seems kind of slow to me based on typical
> > hard disk transfer rates I've seen on storagereview.com and similar
> > sites; I'd expect at least twice this amount.
> >
> > Are my expectations unreasonable? Does Windows have that much of an
> > overhead? Or is there possibly an issue here?
> > Reply here on the newsgroup please.
> >
> > Thanks,
> >
> > Steve
> >
>
>

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

Ok .. here are some questions

Where is your file source that you are copying from? Obviously, you are
copying to your local disks (on an external Storage Vault).

What are your read-ahead and I/O Caching Policy settings for the controller
?

If you are copying over a network, your data I/O will be throttled by the
network. If its from local disks to the storage vault, then yeah ..
something is wrong. Check the policies on both controllers (integrated, and
Perc 4).

I recently put a PERC2 DC, 64bit 128M controller into a mishapen beast I
built. I put 3 Ultra2 drives into a cage that only support WUS3 speeds.
Installed 2003 server on a dual Xeon 2.0 HT box with 3GB Ram. Took 9 minutes
to get through the gui setup.

Once into windows, applied all the patches and updates, updated the PERC FW,
and installed OMAM. Then, I installed HDTach 2.7.

With the default policies for the controller, I was averaging about 16MB/s
sustained I/O. After tweaking the read-ahead and Caching policies, I was
able to achive a very consistent (though still average) sustained I/) of
63MB/s.

Tweak, test, tweak, and test again. Find out what works best on your rig.

- NuTs

"Steve" <swechsler@bigfoot.com> wrote in message
news:1102704995.563057.322760@f14g2000cwb.googlegroups.com...
> Thanks, but I'm quite familiar with different RAID levels. And that
> site neglects to point out that RAID 5 arrays typically take a
> performance hit due to the need to calculate parity for every block
> that's written. While calculating parity doesn't very hard, AFAIK there
> are no controllers extant that can write a RAID 5 array as fast as a
> RAID 0/1 (or RAID 10, if you prefer that nomenclature) array. In the
> database world we generally try to avoid RAID 5 if we have enough disks
> (although many people use them on DSS (decision support [reporting])
> systems where write performance is secondary to maximizing storage
> space).
>
> None of which, of course, answers my original question of whether my
> arrays are functioning slower than they should be.
>
> Steve
>

 

[danr]

Archived from groups: alt.sys.pc-clone.dell (More info?)

Steve, how are the External drives connected (and in what?)

Dan

"Steve" <swechsler@bigfoot.com> wrote in message
news:1102704995.563057.322760@f14g2000cwb.googlegroups.com...
> Thanks, but I'm quite familiar with different RAID levels. And that
> site neglects to point out that RAID 5 arrays typically take a
> performance hit due to the need to calculate parity for every block
> that's written. While calculating parity doesn't very hard, AFAIK there
> are no controllers extant that can write a RAID 5 array as fast as a
> RAID 0/1 (or RAID 10, if you prefer that nomenclature) array. In the
> database world we generally try to avoid RAID 5 if we have enough disks
> (although many people use them on DSS (decision support [reporting])
> systems where write performance is secondary to maximizing storage
> space).
>
> None of which, of course, answers my original question of whether my
> arrays are functioning slower than they should be.
>
> Steve
>

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

it does mention that there is a performance hit with writing to raid 5. why
not break the mirror and test performance without the mirror being active...

"Steve" <swechsler@bigfoot.com> wrote in message
news:1102704995.563057.322760@f14g2000cwb.googlegroups.com...
> Thanks, but I'm quite familiar with different RAID levels. And that
> site neglects to point out that RAID 5 arrays typically take a
> performance hit due to the need to calculate parity for every block
> that's written. While calculating parity doesn't very hard, AFAIK there
> are no controllers extant that can write a RAID 5 array as fast as a
> RAID 0/1 (or RAID 10, if you prefer that nomenclature) array. In the
> database world we generally try to avoid RAID 5 if we have enough disks
> (although many people use them on DSS (decision support [reporting])
> systems where write performance is secondary to maximizing storage
> space).
>
> None of which, of course, answers my original question of whether my
> arrays are functioning slower than they should be.
>
> Steve
>

 

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

OK, back at work, I'm going to try to answer these questions:

Disks are in a Powervault 22XS 14 drive cage (there are actually 14
drives being used, but two are configured in a RAID 1 array and are not
being used for the purposes of this test).

The copy test was done from one local RAID 0+1 array (sorry, my mistake
in calling it 0/1; an old habit) to another. No network was involved.

I am not sure how much cache the controller has, or how it is
configured. It is a PERC4/DC controller. Since the machine's at a
remote datacenter I can't go and look at it. Is there a way to tell the
cache configuration with Dell's OpenManage utilities? Or can it be done
using the controller's BIOS (I can have someone take a look at it)?
Thanks,

Steve

 

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

OK, just figured out that the cache size is 128MB and firmware version
is 3500. No idea how to find out about caching policies, though; I
suspect they're set as the default (whatever that is).

Steve

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

Steve,

I don't have an answer for you but when you find one, could you please
post a followup? I'll be in your situation in the near future and can use
all the tips that I can get.

Thanks,

David

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

"Steve" <swechsler@bigfoot.com> wrote in message
news:1102954556.261527.279070@f14g2000cwb.googlegroups.com...
> OK, just figured out that the cache size is 128MB and firmware version
> is 3500. No idea how to find out about caching policies, though; I
> suspect they're set as the default (whatever that is).
>
> Steve
>

Yer new to this stuff, huh ?

You need to install the Open Manage Aray Manager.

You can TS to the server, and run it locally. Or, you can install the
manager on your workstation, and connect to it remotely (provided your
installed it on the server).

Make sure your firware is the newest. Make sure your driver is the newest.

Dig into the app, and look at the properties of the controller. You will
eventually find what I am talking about.

- NuTs

 

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

I am new to Dell servers, but not RAID in general (in fact, I have an
older LSI controller at home, but it doesn't have the same GUI).

Anyway, I finally found the cache policies (I already had the array
manager installed), and they are as follows (same for all arrays):

Read: Adaptive Read-Ahead
Write: Write-Back
Cache: Direct I/O

I assume these are the defaults.

The drivers are the latest. I haven't been able to determine whether
the firmware is current (the server was built only a couple of months
ago, so it probably is; it's level 3500).

Steve

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

Download HDTach and install it. Run some tests, tweak the policy, and test
again. See what works best with that controller. I already noted the
performance increases i got in another post in this thread.

- NuTs

"Steve" <swechsler@bigfoot.com> wrote in message
news:1103121468.182375.222010@c13g2000cwb.googlegroups.com...
>I am new to Dell servers, but not RAID in general (in fact, I have an
> older LSI controller at home, but it doesn't have the same GUI).
>
> Anyway, I finally found the cache policies (I already had the array
> manager installed), and they are as follows (same for all arrays):
>
> Read: Adaptive Read-Ahead
> Write: Write-Back
> Cache: Direct I/O
>
> I assume these are the defaults.
>
> The drivers are the latest. I haven't been able to determine whether
> the firmware is current (the server was built only a couple of months
> ago, so it probably is; it's level 3500).
>
> Steve
>

 

[steve]

Archived from groups: alt.sys.pc-clone.dell (More info?)

OK, I did some more research, downloaded IOMeter (available free at
sourceforge.net) and discovered that the performance problem lies with
Windows 2000 explorer, not the controller. IOMeter gave me 79 MB/sec
writes and 99 MB/sec reads. This is a vast improvement over the
original numbers.

However, this raises another question. Assuming four drives reading
simultaneously, I believe the read performance should be double what
I'm seeing. Do RAID 0+1 controllers read from all disks, or just one
side of the mirrors (in other words, is the controller smart enough to
grab different blocks off opposite sides of the mirror simultaneously)?
For example, in a four drive array, when I read, am I reading from two
disks or four?

Thanks,

Steve

 

[Guest]

Archived from groups: alt.sys.pc-clone.dell (More info?)

It depends on a couple things. Your volume stripe size, and your NTFS sector
size that was chosen when you formatted the volume.

In a mirror, you should have increased read performance, as you can utilize
2 spindles, since the data exists on 2 physical disks. The controller can
read from both drives simultaneously. Writing is the same as with a single
drive.

However, 79M sustained isnt bad. Just remember, the U spec is what a drive
can burst from the disk to the cache, not necessarily what you will for
sustained I/O on that channel.

- NuTs

"Steve" <swechsler@bigfoot.com> wrote in message
news:1103145431.529776.68830@f14g2000cwb.googlegroups.com...
> OK, I did some more research, downloaded IOMeter (available free at
> sourceforge.net) and discovered that the performance problem lies with
> Windows 2000 explorer, not the controller. IOMeter gave me 79 MB/sec
> writes and 99 MB/sec reads. This is a vast improvement over the
> original numbers.
>
> However, this raises another question. Assuming four drives reading
> simultaneously, I believe the read performance should be double what
> I'm seeing. Do RAID 0+1 controllers read from all disks, or just one
> side of the mirrors (in other words, is the controller smart enough to
> grab different blocks off opposite sides of the mirror simultaneously)?
> For example, in a four drive array, when I read, am I reading from two
> disks or four?
>
> Thanks,
>
> Steve
>