Location of Isoch Resources -- 1394 Fully Managed Case

MARAZAS at BCRVMPC2.VNET.IBM.COM MARAZAS at BCRVMPC2.VNET.IBM.COM
Sat Mar 4 13:22:59 PST 1995


To the IEEE P1394 Interest Group:
===>  Distribution List
===>  P1394 Reflector
===>  X3T10 Reflector
===>  P1394 Trade Association Reflector
===>  This note sent to the X3T10 (SCSI) reflector

This note is a request for comments / reaction to a Bus Management issue
for the draft standard IEEE 1394 -- High Performance Serial Bus. The issue
arises in the context of discussion within the Ballot Resolution Committee
(BRC) regarding comments on the topic arising from the first public review
of draft standard 7.1v1.

The issue concerns location of isochronous resources as applicable to
the following two levels of centralize management defined in the
draft standard:
  <> The Fully Managed level of control
  <> The Limited Management level of control.
The isochronous resources involved are the BANDWIDTH_ALLOCATE CSR
and the CHANNELS_AVAILABLE CSR(s). The INCUMBENT_ANSWER CSR is not
involved in the present discussion.

The present situation as described in draft 7.1v1 has the subject
isochronous resources located at the Full Bus Manager node in the
Fully Managed level of control and at the Resolver node in the case
of the Limited Management level of control. Note, the Resolver node
is also the Limited Bus Manager in the case of the Limited Management
level of control.

In the course of BRC discussions on the topic of Bus Management, there
arose a proposed BRC response in which the subject isochronous resources
would in BOTH LEVELS OF CONTROL be placed at the Resolver node.
This note seeks commentary on the question as to whether the subject
isochronous resources should: (a) be left as presently described location
within draft 7.1v1 or (b) be changed so as to be located at the
Resolver node in both of the levels of control under discussion.
Note, in the Unmanaged level of control there are no such isochronous
resources present and thus there is no issue as to location.

One argument in favor of leaving the isochronous resource location as
per 7.1v1 is that is enables higher levels of bus management services
in the Fully Managed level of control. The point is that resources
located at the Full Bus Manager node are subject to a more intimate and
more extensive level of control by the Full Bus Manager when compared
to the alternative of having these resources at a node likely to be
different from the Full Bus Manager node. An additional point is
that the location of the Full Bus Manager is intended to be fixed
in the presence of bus resets caused by addition and removal of other
nodes. This fixed location for the isochronous resources facilitates
the Full Bus Manager in maintaining a "memory" of isochronous resource
allocations across bus rests. Additionally, the Full Bus Manager
would be more suited to probing "ownership" issues connected
with isochronous resources. Thus, the Full Bus Manager having
the isochronous resources located at its own node is argued to
be in a position to offer more and better level of management
services than if these resources were located at the Resolver node.

Arguments in favor of the proposed change are as follows.
First, there is a more consistent and more convenient access path
to the isochronous resources if they are always at the Resolver
node. The interested parties must determine the location
of the Resolver node in both levels of management control. In
the case of the Fully Managed level of control, there is a level
of indirection involved in the access path. The interested party
must first go to the Resolver node in order to find the location
of the node supporting the Full Bus Manager. In general, the
Full Bus Manager must be assumed to be at a different node than the
Resolver. Thus the interested party must then undertake a second
operation in order to access the node supporting the Full Bus Manager.
It may be observed that the interested party may store the location
of the Full Bus Manager node so that subsequent accesses may be
a one step operation. However, the stored node location of the
Full Bus Manager will become invalid after a subsequent bus reset.

Another argument in favor of the proposal is the timeliness with which
an access to the isochronous resources may be made. The identity of
the Resolver node is known immediately as a consequence of completing
the Tree_ID and Self_ID process. In the proposal, the location of
the isochronous resources is decoupled from the question of determining
the bus management entity. In the draft 7.1v1, access to the
isochronous resources may only take place once the identity of
the bus management entity is determined. Given that the protocol
for determination of the management entity is relatively long, one
second after reset, there is the need per draft 7.1 to wait this time interval
in both the Full Bus Management and also the Limited Bus Management
levels of control. Some individuals prefer the more rapid access to
the isochronous resources. Other individuals prefer the wait time
as it allows the Full Bus Manager entity to have greater control
over the access procedure to the subject isochronous resources.

Please comment and state your preference along with your reasons.

Best regards,
Jerry Marazas,
P1394 Chair




More information about the T10 mailing list