
IBM2520L8767
IBM Processor for ATM Resources
Transmit Cell Scheduler (CSKED)
Page 228 of 553
atmrm.chapt05.01
08/27/99
Bit(s)
Name
Description
21
Disable overlapping transmit requests
This bit is meant for debug purposes only. It will disable the ability of CSKED
to overlap requests to SEGBF.
20
Close connection immediately
Setting this bit will cause segmentation to stop immediately on any connection
that has been issued a close connection primitive. If this bit is not set an event
will be generated after all traffic queued to this connection has been sent.
19
Queue LCD based memory management
events
Setting this bit will cause LCD based memory management events to be
queued to the Transmit Complete Queue, if enabled by bit 18 of this control
register. If LCD based memory management is enabled and this bit is off, the
receive pool ID associated with this connection will be updated when a thresh-
old is crossed.
18
LCD based memory management
Setting this bit will enable LCD based memory management for received
packets. See
Logical Channel Data Structure
on page 38 and
Definition of
LCD-Based Memory Management of Transmit LCD
on page 45 for further
information on this function.
17
Use bandwidth scheduling for low priority
Setting this bit will cause low priority traffic to be bandwidth scheduled. The
peak interval in the LCD is used to provide a relative weight in determining the
amount of bandwidth the connection will use. For example a peak interval of
one will use twice the bandwidth as a connection with a peak interval of two.
The average interval specifies the maximum rate that the connection can use.
For example if the average interval is set to ‘2’, the maximum rate at which it
can send a cell is every two timeslot times (as defined in the Timeslot Pres-
caler Register).
16
Control scheduling of medium priority traf-
fic with GFC reset for group A
Setting this bit will cause medium priority traffic to be controlled by the group A
GFC bits in the ATM header of received cells.
15
Control scheduling of low priority traffic
with GFC reset for group B
Setting this bit will cause low priority traffic to be controlled by the group B
GFC bits in the ATM header of received cells.
14
Queue the LCD address if freeing and
queueing
Setting this bit will cause the LCD address, instead of the packet address, to
be queued if both freeing and queueing on transmit complete.
13
Disable virtual buffer error detection
Setting this bit will cause the buffer enqueue logic to ignore virtual buffer
errors.
12
Disable queuing virtual buffer errors
If virtual buffer error detection is not disabled, detected errors will be queued.
If this bit is set, this queuing is disabled and the buffer will be freed.
11
Reserved
Reserved.
10-8
SEGBF Queue Length Threshold
Cells can be queued in SEGBF up to the number specified in this register.
The default is seven which is above the limit for pass two. Writing these bits to
zero will also disable this function.
7
Disable SEGBF Queue Length in Schedul-
ing
CSKED will normally include SEGBFs queue length in the calculations when
rescheduling a cell. If this bit is on it will disable this function and the cell will
be scheduled as if the cells were transferred when SEGBF accepted the cells.
6
Priority of buffer requests
If this bit is not set, scheduling requests have a higher priority than buffer
requests. If this bit is set this priority is reversed. It should be set if a significant
percentage of packets are only a few cells in length.
5
Enable timers
Timer descriptors can be enqueued to this entity that will cause a DMA
descriptor to be executed on expiration. If these timers are used, this bit must
be set. If they are not used, this bit should be reset.
4
Reschedule Packets in the Slow Queue to
the Fast Queue
This function is not implemented in pass one. It is implemented in pass two.
If the average or peak interval is greater than 255, the cells will be scheduled
in the slow queue. The slow queues will be serviced every 64 prescaler time
units. This means that a jitter of up to 64 prescaler time units should be
expected for slow traffic. If this bit is set, packets in the slow queue will be
rescheduled at the appropriate time to the fast queue. This will decrease the
variation in the scheduling but may cause some performance degradation if
traffic is heavy.