Re: iSCSI: MaxBurstSize

["John Hufferd" <hufferd@us.ibm.com>]

I believe, we need to consider this as issue with devices that can be both
initiators and targets.

Take for example a streaming tape; when it is reading the data from a Disk
Storage Controller.  Often these device start a long I/O streaming action
and have less memory then will contain the data they are requesting.
Through various double buffering techniques, they can keep the data coming
in, and going out to tape, without back-hitching.

So I believe MaxBurstSize has value, even if the disk storage folks think
it is minimal.

.
.
.
John L. Hufferd
Senior Technical Staff Member (STSM)
IBM/SSG San Jose Ca
Main Office (408) 256-0403, Tie: 276-0403,  eFax: (408) 904-4688
Home Office (408) 997-6136, Cell: (408) 499-9702
Internet address: hufferd@us.ibm.com


Santosh Rao <santoshr@cup.hp.com>@ece.cmu.edu on 02/21/2002 01:35:49 PM

Sent by:    owner-ips@ece.cmu.edu


To:    ips@ece.cmu.edu
cc:
Subject:    Re: iSCSI:  MaxBurstSize



Pat,

The MaxBurstSize login key is a don't care for iscsi initiators. It
serves mostly an informational purpose, letting the initiator know at
login time as to what is the maximum amount of data it can expect the
target to request in 1 R2T request, or the maximum amount of data it can
receive from the target in 1 data-in sequence.

Most initiators should be negotiating this to the maximum possible
value, thereby, indicating to the target to just return the maximum
value it supports for MaxBurstSize.

The MaxBurstSize itself is inherited as one of the tunables from the
scsi Disconnect-Reconnect Mode Page as defined in SPC-2, which is to be
re-defined for each SCSI transport as appropriate.

IMO, the "MAXIMUM BURST SIZE" tunable of the disconnect-reconnect mode
page makes sense in the parallel scsi context alone, where, either side
(initiator or target) can specify how long the scsi bus can remain in
the data phase before a disconnect must be done.

In a fabric based SCSI transport, where, the data transfer does not
cause any bus to be tied down, and the initiator's data buffers are
allocated up-front before the SCSI command is sent to the target, I
don't see any practical use of this key for initiators other than being
informational, and allowing them to learn of the target's limits
up-front at login time.

I don't see why iSCSI needs to define this key, if its informational
utility is of little or no value. I am also curious about what value
FCP-2 initiators would find in this tunable. (?)

Regards,
Santosh



"THALER,PAT (A-Roseville,ex1)" wrote:
>
> Julian,
>
> I still don't see the rational for MaxBurstSize.
>
> The initiator doesn't need it to limit its required resources. The
initiator
> has to have its buffers allocated when it issues a command. It has to
> control the demand on its resources by limiting the commands it issues
based
> upon its available resources.
>
> The target has FirstBurstSize and R2T's to control the demand on its
> resources.
>
> The other use of MaxBurstSize is to limit how often target may set the A
> bit, but the utility of this is unclear.
>
> Regards,
> Pat Thaler
>
> ---Original message----
> From: Julian Satran [mailto:Julian_Satran@il.ibm.com]
> Sent: Tuesday, February 12, 2002 12:33 AM
> To: Robert D. Russell
> Cc: ips@ece.cmu.edu
> Subject: Re: UNH Plugfest
> +++
>
> <Section 3.8.3 should be 10.8.3>
> 3. Section 3.8.3 limits the value of the Desired Data Transfer Length in
>   an R2T to at most MaxBurstSize.  What is the rationale for this?
>    An R2T is sent by the target to the initiator, so why can't the
>   target specify any size it wants in the R2T?  The target already
>   uses R2Ts to control the flow of Data-Out PDUs from the initiator,
>   so why impose this restriction on the R2Ts?
>
>   Could someone please explain the benefit to this limitation on R2Ts?
>
> +++ Is this a plugfest question or one of your own?  For your own
questions
> the channels are always open.  The MaxBurstSize is there to enable the
> initiator to share resources between several executing commands and limit
> the number of "pending buffers" a target will have to keep
> in case one of the Data Out PDUS is damaged and transfer to a device is
not
> possible.
>
> +++

--
##################################
Santosh Rao
Software Design Engineer,
HP-UX iSCSI Driver Team,
Hewlett Packard, Cupertino.
email : santoshr@cup.hp.com
Phone : 408-447-3751
##################################