X-MimeOLE: Produced By Microsoft Exchange V6.5
Received: by onstor-exch02.onstor.net 
	id <01C82BB3.633AC859@onstor-exch02.onstor.net>; Tue, 20 Nov 2007 12:24:44 -0800
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
Content-class: urn:content-classes:message
Subject: RE: Follow up question about Max Queue Depth
Date: Tue, 20 Nov 2007 12:24:44 -0800
Message-ID: <BB375AF679D4A34E9CA8DFA650E2B04E06648D71@onstor-exch02.onstor.net>
In-Reply-To: <018001c82bb2$7257f840$0200a8c0@lab.css.glasshouse.com>
X-MS-Has-Attach: 
X-MS-TNEF-Correlator: 
Thread-Topic: Follow up question about Max Queue Depth
Thread-Index: AcgrsnB8L5KWLLpcQeyzQ9PCw7EOwQAABMXw
From: "Bill Nadzam" <bill.nadzam@onstor.com>
To: "Fred McFadden (Glasshouse)" <fredm@css.glasshouse.com>,
	"dl-Customer-Engineering" <dl-Customer-Engineering@onstor.com>
Cc: "dl-cstech" <dl-cstech@onstor.com>

Not sure what he means about High Water Mark?

However; the question they asked today was how to alter the number of
I/O's to a Lun.
This is done by editing the sdm_devcap file. "/onstor/conf/sdm_devcap"

Locate the Controller type entry in the file and edit the
MODEL_MAXDISPATCHCNT to the value.
If this entry does not exist, just add it in the correct model field.


i.e.

#######################################################################
## ARRAY: HITACHI 9500V Series
##
## Hitachi arrays with the product id DF600
##
#######################################################################
[DISK_START]
        VENDOR        =3D HITACHI
        PRODUCTID     =3D DF600F
        REVISION      =3D *
        STATE_MACHINE =3D HITACHI-9XXX

        [MODEL]
                MODEL_KEY_STR =3D D60
                MODEL_DISPLAY =3D HDS9500V
                MODEL_MAXDISPATCHCNT =3D 10
        [MODEL]
                MODEL_KEY_STR =3D *
                MODEL_DISPLAY =3D HDS_DF600F
                MODEL_MAXDISPATCHCNT =3D 10
[DISK_DONE]
=20

-----Original Message-----
From: Fred McFadden (Glasshouse)=20
Sent: Tuesday, November 20, 2007 12:18 PM
To: dl-Customer-Engineering
Cc: dl-cstech
Subject: Follow up question about Max Queue Depth

Any suggestions on a response for Customer (Ithaka) appreciated:

Customer now asking about I/O high water marks in line with this
question about Max Queue depth.=20
I suspect we can not hit the theoretical high water mark.
We do have evm stats command to measure i/o.=20
I don't know how one would define how to report a "you've reached the
limit"
indication.

Thanks
Fred

----------------customer latest response---------------------- Blue
Thunder Somogyi commented:

In response to Engineering's statement "In practice this does not really
happen" and the following explanation, this does not seem to be a
logically consistent argument. In the case of a 1TB volume consisting of
ten 100GB LUNs, on which there are thousands of simultaneous accesses of
data scattered across the volume, I do not see how the described file
allocation pattern prevents there from (potentially) being 100
outstanding IO requests at a given time.

More to the point, does Onstor have any tool, or is it possible in
EverON OS, to report on IO high water marks to determine whether this is
indeed the case (ie IO queues reaching close to theoretical maximum).

--BTS


----------------Response from Bill Nazdam-------------
From: "Bill Nadzam" <bill.nadzam@onstor.com>
To: "Alan Cooke (Glasshouse)" <acooke@css.glasshouse.com>; "dl-cstech"
<dl-cstech@onstor.com>
Sent: Friday, November 16, 2007 12:53 PM
Subject: RE: Case 6450 - Question about "maximum queue depth setting"
case 6450

The Max Queue Depth is on a LUN.

If a Volume is comprised of 10 luns and the I/O was spread equal across
all the luns.
Then each lun would have 10 outstanding I/O's each.

In practice this does not really happen.

A Volume is a Linear concatenation of extents ..

It appears to the file system as one large number of linear blocks.

The file system allocator spreads the usage of those blocks across the
linear space.
In effect files are spread across the blocks in a random usage.

Sequential I/O is different, and allocations tend to be sequntial across
the linear space.
=09

-----------Opening case notes from Ithaka Case 6450---------------------
I\'ve found in the documentation that there is a max queue depth setting
on a per LUN basis, but I would like to find out if there is a maximum
queue depth setting for a given storage array. This would be the maximum
number of in-flight IO total for all LUNs owned by a given storage
array, regardless of number of LUNs configured from that array on that
Bobcat. If this setting does exist, I would like to know what its
default value is, and whether it is adjustable.


