Re: Disk Queue Length

[Michael_Smith@xxxxxxxxxxxxxx]

message: 1
date: Wed, 2 May 2007 12:01:59 -0500
from: Alexei Pytel <pytel@xxxxxxxxxx>
subject: Re: Disk Queue Length

> > I have a 2757 controller in the 810 with six (6) 4326 (36 GB) drive in
RAID 5 array.  In the 7116 expansion cage, >> I added five (5) 4327 (70GB)
drives. I have the six 4326 drives plus the last 4327 in ASP1.  I have the
first
> > five 4327 drives in ASP3.

> This makes six 4327 drives. Above you've said you have five.

Sorry about that. You are correct.
ASP 1 : 6 36 drives + 1 70 GB drive.  ASP 3 : 4 70 GB drives.
RAID 5 array 1 : 6 36 GB drives (in main 810).
RAID 5 array 2 : 5 70 GB drives (in 7116 expansion unit).

> > My first question is :  Is the Total Queue Elements column showing all
the I/O request to that drive over the 15
> > minute interval? The number of samples taken in the 15 minute interval
is 1786 (about 2 a second).  My
> > typical values for Total Queue Elements range from 50-90 for the 15
minute interval.  Am I to assume that would
> > translate into slightly more than 2 queue elements every 30 seconds?
(70 TQE/15 min = 4,67/2 samples per
> > minute = 2.3)

> "Total Queue Elements" is a bit of a misnomer - it shows how many
operations were oberved at the time of sampling. > This includes the
operations waiting in a queue and opertion which is currnetly in progress
(if any).
> By dividing "Total Queue Elements" (DSQUEL) by number of samples (DSSMPL)
you will get an estimate of average
> number of operations for this disk unit at any given time.
> This is described in InfoCenter articale:

>
http://publib.boulder.ibm.com/infocenter/iseries/v5r4/index.jsp?topic=/rzahx/rzahxqapmdisk.htm


> In your case, if you have 90 for QSQUEL for a 15 minute interval, it
means that at any given time you had about
> 90/1786 = 0.05 operations in progress - very low indeed. The total number
of operations processed by this disk
> unit over the interval is DSRDS + DSWRS - reads plus writes (and this is
a fact - not an estimate).

Thanks, that's some reat infomation.


> > Second question:  Like I said, I have one 70 GB drive in ASP 1 and the
rest are all 36 GB drive.  I notice the 70 >> GB drive has a higher Total
Queue Element value.  It would make sense that since it's a larger drive,
there's
> > more data on it, so more request for read/write.  Since i5/OS spreads
data over all available disk units in ASP
> > proportionally to the size of the disk units, in average a disk of a
double size will have double operations
> > targeted to it.

> Since technical capabilities of disk units of the same technology are
about the same regardless of their size,
> this disk unit will have to work that much harder.  Because of this, if
you have more than moderate disk I/O load
> in the ASP, it is not a good idea to mix disk units of significantly
different sizes in the same ASP.

> > Does this drive have to handle request from both ASP's.  The 'standard'
read/write from OS/400 and
> > applications in ASP1 and the checksum at the RAID level for ASP3?

> Yes, in general disk units in the RAID set have to handle parity
management operations for other disk units in the > same RAID set - even if
they are in the same ASP. (Storage adapter which handles RAID, has no
awareness of ASPs.) > This varies depending on how parity data is located
in disk units.

> > I thought the checksums on the IBM RAID array was contained on the first
four (4) drives of an array.  So, ASP 1
> > has the OS/400 stuff on the first four and ASP 3 has the Domino stuff on
it's first four.

> You can tell at a  glance, which disk units have parity data on them by
looking at their capacity - capacity of
> the units with parity is smaller. To determine this more reliably, look
at disk model number:
> 030: Unprotected or mirrored unit attached to a non-RAID capable
controller.
> 050: Unprotected or mirrored unit attached to a RAID capable controller.
> 070: Non-parity member of a parity (RAID) set. Full capacity.
> 071: Parity member of a parity (RAID) set with 16 parity members.
Fifteen-sixteenths capacity.
> 072: Parity member of a parity (RAID) set with eight parity members.
Seven-eighths capacity.
> 074: Parity member of a parity (RAID) set with four parity members.
Three-fourths capacity.
> 078: Parity member of a parity (RAID) set with two parity members. Half
capacity.


My disk are usually more active than this.  It's a really nice day in
Albany for a change...
The parking lot is sparse.. not much going on.

Disks 1-6 + 17 are in ASP1
Disk 13-16 are in ASP3.

Since 17 is at full capacity, it does not contain RAID info.  As a result
it us just handling ASP1 request.
               Size Model   %     I/O   Request   Read  Write   Read  Write
%
Unit  Type      (M)       Used    Rqs  Size (K)    Rqs   Rqs     (K)   (K)
Busy
   1  4326    26373  074  77.4     3.9      7.6      .9    2.9    9.3
7.1     0
   2  4326    26373  074  77.4    12.1      4.7    10.2    1.9    5.1
2.6    33
   3  4326    26373  070  77.4    18.1     10.6    12.5    5.5    9.5
12.9     0
   4  4326    35165  074  77.4    15.1      4.4     1.6   13.4    6.4
4.1     0
   5  4326    26373  074  77.4    17.7      4.0    13.4    4.2    4.5
2.4    16
   6  4326    35165  070  77.4    22.3      5.2     7.2   15.1    6.9
4.4    16
  17  4327    70564  070  77.4   110.5      5.5    67.8   42.7    4.9
6.5    33
  13  4327    52923  074  34.8    62.8      9.6    21.3   41.4   16.1
6.2    14
  14  4327    52923  074  34.8    66.8     12.4    35.5   31.2   17.0
7.2    16
  15  4327    52923  074  34.8    74.0     13.7    41.1   32.9   18.1
8.2    16
  16  4327    52923  074  34.8    60.5     11.4    25.0   35.5   14.2
9.4     0


Thanks for the clarification.


Michael Smith
Technical Support and Data Center Manager
Farm Family Insurance cos.