Figure 1 shows the baseline performance of an ATM VIP 2 CAR when mangaging 80 CAR instances at a time: CPU saturation is already reached at only 6000 pack/sec.
Fig.1: CPU utilization on an ATM VIP2 interfaces with 80 CARs, 1-entry ACL per CAR, 1 match for each CAR instance
By reducing the number of streams to 1 (the number of CAR instances is still 80), we studied the impact of the matched CAR position on CPU. As Figure 2 shows, the inpact is considerable, for any packet rate. We can say that the match position together with the packet rate is one of the primary factors which determines the CAR overhead.
Fig.2: relationship between CPU overhead and match position in an ACL when using CAR.
In our third test, we reduced the number of CAR instances to 1 and increased the
ACL length to 80. In each test we used a single stream and we progressively increased
the match position within the access list.
As Figure 3 shows, with a rate of 9000 pack/sec, even in the worst
case (match position=80) the CPU utilization is below 50%, this seems to suggest that
the match position within an ACL list has a smaller impact that the CAR-matched position.
Figure 4 displays the same results by emphasizing the impact of
the match position .
Fig.3: CPU utilization vs packet rate for different match positions
Fig.4: CPU utilization vs packet match position