Feedback solicited - read performance is awful with PE2650runningRH AS

Wayde Nie niew at mcmaster.ca
Tue Oct 28 16:03:00 CST 2003


On Tue, 2003-10-28 at 06:12, Gary Mansell wrote:
> Thanks for your response - I cannot immediately see why your performance
> should be OK when mine is problematic (I have tried this on two
> identical PE2650's) This problem has also been seen on a HP system.
> 
> Perhaps you can have a look at the bug #106771 report on
> http://bugzilla.redhat.com as I would be interested to hear if you can
> think of any reason why I am seeing the problem but you are not

Hi Gary,

I seem to have caught what you've got... I've scanned the bugzilla
report you quoted and nothing catches my eye off hand... I'll look
closer when I get a chance.

I'm not sure what's changed here wrt configuration (this box is a test
bed for ad hoc things I might need to work on), but write has improved
considerably (~39Mb/s), read has remained pretty much the same
(~22Mb/s).. It seems odd that write should be so much better than read
on a HW Mirror... 

The typical output of a dd run from today's tests were:

/usr/bin/time dd if=/dev/zero of=./testfile bs=16384 count=250000 
250000+0 records in
250000+0 records out
0.29user 32.11system 1:35.82elapsed 33%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (137major+17minor)pagefaults 0swaps
(About: 39Mb/s on a 4096000000 byte file)

/usr/bin/time dd if=./testfile of=/dev/null bs=16384
250000+0 records in
250000+0 records out
0.24user 12.94system 3:12.64elapsed 6%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (137major+19minor)pagefaults 0swaps
(About: 22Mb/s on a 4096000000 byte file)

On a side note, what is the generally expected rates for these PERC3Di's
if things are otherwise healthy?

To fill in some data points for this box, it has System Bios A10, ESM
fw:2.10, Pri and Sec Backplane fw: both 1.01. The PERC3Di BIOS is:
2.7-1[Build 3170].

There is also an addon Adaptec39160 card with no devices attached (BIOS:
v2.57.2S2).

All four disks on the PERC3Di are in the "hot-swap" bays with each
mirror-set defined across the 3Di channels.

I'm also using an LVM logical volume for the test partition. It's a
stripe across the two pv's (sda/sdb - PERC hw mirrors).

At this point does the trouble seem localized to RHAS on PE2650's (with
PERC 3Di's)? Should I (rather, am I likely to ;) experience any problems
upgrading a pair of RHL7.3 PE2600 (PERC4Di's) that I have to RHAS3?

> ----------------------------------------------------------------- <
> Wayde Nie                                                         <
> Software Analyst               Computing and Information Services <
> phone: (905)525-9140 ext 23856 McMaster University, CANADA        <
> fax:   (905)524-5288           niew at mcmaster.ca                   <
> ----------------------[\]--->=\o.:---[\]------------------------- <


> On Tue, 2003-10-21 at 22:07, Wayde Nie wrote:
> > Hi,
> > 
> > I've got a PE2650 running RHAS3b2. It has 4-36Gb disks in 2 HW mirror
> > sets (presented as sda and sdb) on a PERC3Di (Adaptec aacraid driver).
> > The box has 2x 2.4GHz Xeon's with HT enabled and 2Gb RAM.
> > 
> > The dd test results as follows:
> > 
> > [root at feweb tmp]# /usr/bin/time dd if=/dev/zero of=./testfile bs=16384
> > count=250000
> > 250000+0 records in
> > 250000+0 records out
> > 0.29user 47.06system 6:35.95elapsed 11%CPU (0avgtext+0avgdata
> > 0maxresident)k
> > 0inputs+0outputs (138major+15minor)pagefaults 0swaps
> > ~10.34 Mb/s Write rate.
> > 
> > [root at feweb tmp]# /usr/bin/time dd if=./testfile of=/dev/null bs=16384
> > 250000+0 records in
> > 250000+0 records out
> > 0.21user 17.91system 3:03.71elapsed 9%CPU (0avgtext+0avgdata
> > 0maxresident)k
> > 0inputs+0outputs (137major+18minor)pagefaults 0swaps
> > [root at feweb tmp]#
> > ~22.26 Mb/s Read rate.
> > 
> > > ----------------------------------------------------------------- <
> > > Wayde Nie                                                         <
> > > Software Analyst               Computing and Information Services <
> > > phone: (905)525-9140 ext 23856 McMaster University, CANADA        <
> > > fax:   (905)524-5288           niew at mcmaster.ca                   <
> > > ----------------------[\]--->=\o.:---[\]------------------------- <
> > 
> > On Tue, 2003-10-21 at 07:17, Gary Mansell wrote:
> > > I would really appreciate some help from the list here to see if anyone
> > > else is suffering from the same problem that I am. I appear to have
> > > fallen foul of a problem with RH AS products running on Dell PE2650's.
> > > 
> > > I have raised bug number #106771 at http://bugzilla.redhat.com and I
> > > have tried to raise the issue with Dell but I am not getting very far -
> > > both parties blame the other and I am stuck in the middle getting
> > > nowhere.
> > > 
> > > I have installed RH AS 2.1 and RH AS 3.0b2 and have found that the read
> > > performance from the system mirror is awful - it is less than the write
> > > speed!! If I install RH 9.0 or RH 7.3 the read performance to this same
> > > device is fine.
> > > 
> > > Please could anyone who has a Dell PE2650 with a version of RH AS
> > > installed run the following quick and dirty (but representative) disk
> > > performance tests and let me know if they find the same.
> > > 
> > > Write Performance:
> > > /usr/bin/time dd if=/dev/zero of=/testfile bs=16384 count=250000
> > > 
> > > Read Performance:
> > > /usr/bin/time dd if=/testfile of=/dev/null bs=16384
> > > 
> > > 
> > > FYI, I have a Dell PE2650 with 2x 2.4GHz Zeons (Hyperthreading enabled),
> > > 2GB RAM, 2x18GB internal system disks in a HW mirror configuration using
> > > the Adaptec PERC 3Di.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.us.dell.com/pipermail/linux-poweredge/attachments/20031028/fdb9e64f/attachment.bin


More information about the Linux-PowerEdge mailing list