home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
OS/2 Shareware BBS: 8 Other
/
08-Other.zip
/
SDLCPO.ZIP
/
SDLCPOLL.TXT
Wrap
Text File
|
1992-11-30
|
5KB
|
114 lines
02/06/92
TITLE: OS/2 EXTENDED EDITION - PU2.1 SDLC
SYMPTOM:
Inability to activate a session to an OS/2 Communications Manager
workstation configured to function as a PU2.1.
Message log reflects an ACS0119C entry.
ENVIRONMENT:
PS/2 - All microchannel models
Host Communications Adapter - Multiprotocol Adapter
Symptom appears when a SDLC link experiences polling delays caused by one
or more of the following:
o Slow link speed
o Too many PUs on link
o Communication errors on link
o Excessive link utilization
o Excessive network delays
CIRCUMVENTIONS:
o Increase line speed.
o Reduce number of PUs on link.
o Repeat PU2.1 addresses in 'Service Order Table' of NCP.
o Eliminate communication errors on link.
o Tune network
o Add '#Y' parameter to the SDLCDD.SYS statement in CONFIG.SYS.
ANALYSIS:
The SDLCDD.SYS driver has a hard coded value for the 'Inactivity Timer.'
DETAIL:
The '#Y' OS/2 SDLC device driver parameter is provided by OS/2 EE 1.3 and
ES Communications Manager products as a means for alleviating some of the
layer 2 performance problems encountered when activating PU2.1 nodes on a
SDLC multi-point line. The following is an explanation of how this anomaly
occurs:
o Before PU2.1 nodes enter a 'Contacted' state they must go through a
series of XID exchanges with the PU4/5 node for PU identification and
parameter negotiation. This overhead does not occur with PU2.0 nodes;
they directly enter a 'Contacted' state with a simple SNRM-UA exchange.
o The aforementioned XID overhead is not a problem in and of itself.
Link delays aggravate the problem by causing XID exchanges to go
through several 'cycles' before the nodes are ready to enter a
'Contacted' state. Since all of the workstation nodes are Secondary
they are powerless to access the line until they are given a poll
response opportunity.
o Before a SDLC node enters Normal Response Mode (NRM, this is the
'Contacted' state) it is in Normal Disconnected Mode (NDM). NDM is the
equivalent of 'chaos' mode since there are no architected state machine
rules concerning timing for commands and responses. The OS/2 SDLC
device driver does not generate XID responses; it waits for the
Configuration Services portion of the PU layer to build and supply the
XID responses. To ensure the link does not timeout while waiting for
the XID from Configuration Services, the SDLC driver will intentionally
wait for the next poll prior to sending the XID response. This method
works fine except when there is excessive link polling delays which
results in the secondary node 'timing out' due to inactivity.
The primary focus for creating an efficient multi-point environment should
be to balance the number of PUs per line with the amount of available
bandwidth (line speed). In addition to this, the '#Y' parameter provides
two additional benefits:
1. The SDLC Inactivity Timer gets set to 80 seconds. This prevents the
node from 'timing-out' during XID exchanges on heavily loaded or slow
multi-point lines. Normally, the inactivity timer is internally set to
40 seconds for most configurations.
2. The SDLC driver will wait for Configuration Services to provide the
XID, rather than waiting for the next poll. This reduces the number of
poll cycles it takes for a node to get into a 'Contacted' state since
the node does not wait for repolls. The '#Y' parm has a symbolic
meaning: the '#' stands for 'collision risk' and the 'Y' (or 'y') means
'yes' thus yielding 'collision risk = yes'; though this significantly
reduces node activation time, users of this parameter need to read and
understand the following concerns.
CONCERNS:
o The '#Y' parameter is not an all-encompassing fix for PU2.1 multi-point
environments. As previously stated, this parm, used in conjunction with
number-of-PUs-to-line speed ratio balancing, significantly alleviates
node activation performance problems at layer 2 (SDLC). This parameter
in no way obviates potential performance problems that may or may not
be encountered at layer(s) 3+ (LU activation, etc.).
o Usage of this parameter means that the user, or network administrator,
assumes responsibility for ensuring that the poll timeout value of
primary SDLC node (usually a controller) is sufficient to allow the
workstation to issue the XID response. A poll timeout value of 1 second
or greater is strongly recommended. Specifically this is the REPLYTO
statement in the GROUP definitions of NCP, the default is one second.