PM8611-BIAP PMC [PMC-Sierra, Inc], PM8611-BIAP Datasheet - Page 267

• Manufacturer Part Number: PM8611-BIAP
• Manufacturer: PMC [PMC-Sierra, Inc]

13.3.5
13.3.6
Proprietary and Confidential to PMC-Sierra, Inc., and for its Customers’ Internal Use
Document ID: PMC-2010883, Issue 2
There are 3 maximal length paths in the merged graph, (A 2 B 1), (D 5), and
(C 4 F 6 E 3). The last path mentioned requires re-labeling. If we start with edge
(C 4) and traverse the path, alternately labeling with timeslot 2 and 3, we get the graph in
Figure 28. The timeslot labeling in this graph replaces timeslots 2 and 3 in the original graph
(and schedule).
Figure 28 Relabeled Graph
Experimental Results
The performance of PMC-Sierra’s Open Path Algorithm has been studied by implementing it in
C++ and running extensive random connection tests. Tests for NSE/SBSLITE applications of this
algorithm used a single NSE-20G connected to 32 SBSLITEs, each carrying a full complement of
DS0 connections (258,048 DS0 calls). Many runs were completed in which an initially unloaded
switch is presented with a sequence of random call establishment requests up to the point of
100% switching loads. These runs were carried out on a 600 MHz Alpha running Linux. In all of
these runs, no otctet open path search took longer than 10 s, thus supporting up to 100,000
call establishments per second. T1s and other aggregates require the establishment of multiple
octet open paths; complete T1s can be established at about 3,700 T1/sec. The reasons for this
surprisingly good performance are explained in the separate document describing the Open Path
Algorithm. It is our opinion that these rates are sufficiently high that the call establishment
algorithm should not be a bottleneck in any application of the NSE/SBSLITE, and that this rate is
sufficiently high to permit the NSE/SBSLITE to be used for PSTN call establishment rates (up to
100,000 calls/sec in a switch supporting 258,048 full-duplex calls, with the switching core
implemented in 1 NSE-20G and 32 SBSLITE chips).
Multicast
Scheduling general multicast connections is an entirely different class of problem. With
unrestricted multicast, the underlying architecture is non-blocking up to capacity dictated by the
number of slots in a frame, but finding the non-blocking schedule is NP-hard. There is no
polynomial time running algorithm known to solve this class of problem.
1
This ignores inband or µP to NSE limitations.
A
1
B
2
C
3
D
4
E
5
SBSLITE™ Telecom Standard Product Data Sheet
F
6
Preliminary
1
DS0
266