|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: iscsi : DataPDULength can differ in each direction.
Rod,
Actually you bring up a valid point. The length different in the two directions is perhaps the less important feature in the floated proposal. However it stressed (once again) that we need a more flexible PDUlength scheme:
- that should allow dynamic changes (some of the framing schemes proposed will require that)
- that can be different on different connections
Having it different on the two directions is a slight side effect that can be beneficial and comes at zero price once you have done the former.
And the I think that it can be done with minimal disruption (very little code and this only if you implement ErrorLevel 1 and 2).
Julo
| "Rod Harrison" <rod.harrison@windriver.com>
Sent by: owner-ips@ece.cmu.edu
07-10-01 22:22
Please respond to "Rod Harrison"
|
To: Julian Satran/Haifa/IBM@IBMIL, <ips@ece.cmu.edu>
cc:
Subject: RE: iscsi : DataPDULength can differ in each direction.
|
Julian and all,
Before you go ahead and spend time and energy on this I would just
like to re-iterate my concerns that in practice few, if any,
implementations will support asymmetric PDU sizes.
If, for example, a software initiator indicates a max PDU size of
128k and a hardware target indicates 8k, do we really think the target
will ever send a 128k PDU? I don't believe it will. The supporting
argument seems to be predicated on the fact that the buffer management
changes needed to make this happen are trivial. However, if an
implementation can chain buffers together on transmit to build PDUs
larger than the maximum size offered why can't it do so on receive?
And if it can do so on receive why would it ever offer a maximum lower
than the real maximum it can support?
As an aside, I think this change might create something of a headache
for sniffers since they now have to buffer the larger of the offered
sizes instead of the smaller.
I agree that asymmetric PDU sizes seem beneficial but I have a very
hard time believing we'll see any implementations that can take
advantage of this feature. I also think the benefit might be
relatively small and is probably not worth the level of spec change at
this late stage.
- Rod
-----Original Message-----
From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of
Julian Satran
Sent: Saturday, October 06, 2001 8:04 PM
To: ips@ece.cmu.edu
Subject: RE: iscsi : DataPDULength can differ in each direction.
As I said this requested change has value and we are going to do our
best to incorporate it in 09.
Fortunately PDUlength has few dependencies in the protocol.
Unfortunately (as I said already) we would like PDUlength to a be a
per connection parameter as it is it closely related to the Path MTU.
This later assumption may require some changes in the recovery
mechanism (the current recovery mechanism assumes that any PDU can be
"replayed" as it was on every connection and this assumption won't
hold anymore).
Mallikarjun and myself will attempt to get a better understanding of
the things required and will keep you updated.
I think we have enough information to work on it and we will keep the
list updated on what we think can be done and how.
We have several alternatives and would appreciate some timeout on this
thread.
Julo
Home
Last updated: Mon Oct 08 15:17:29 2001
7128 messages in chronological order
|