Table of Contents
Advertisement
Chapter 2
Scalability and Performance
Queue Scaling
The Queue Scaling feature increases the total number of queues that VTMS supports to 131,072. Of the
total number, 254 queues are available for high speed interfaces, and 130,816 queues are available for
low speed interfaces. This increase allows the support of the 31,500 priority queues (of 131,072 total
queues) on 31,500 sessions or interfaces.
Each interface includes a class-default queue and a system queue. If you attach an output policy map
with 1 priority queue and 1 class-based weighted fair queue (PQ/CBWFQ) to each of the 31,500
interfaces, the number of priority queues is 31,500 and the total number of queues is 31,500 x 4, or
126,000 queues.
The maximum number of queues per link remains at 32, of which 29 are user-configurable because there
is 1 class-default queue, 1 system queue, and 1 reserved queue.
To support 131,072 queues, the queue limits range has changed. For high-speed interfaces (an interface
that has a speed greater than 622 Mbps), the queue limit range is 128 to 65,536. For low-speed interfaces
the queue limit range is 8 to 4,096. Because the total number of packet buffers for queue limits is
4,194,304, the average queue depth is less than or equal to 32 per queue with 131,072 queues configured.
On low-speed interfaces, the default queue size is 8 for all QoS CBWFQ queues, with the exception of
WRED queues. The default queue size for WRED queues is 32.
The class-default queue size on low-speed interfaces has changed from 32 to 8. If the traffic is too bursty
and packets drop, you can use the queue-limit command to increase the class-default queue size.
If you change the queue size for 131,072 queues while traffic is running, the queue size for a few queues
might not be changed if packets were in the queues. An "out of resource" message can also appear. Use
the queue-limit command to modify the queue size for those queues that were not changed.
The queue limits packet buffers can become fragmented after the queue sizes on 131,072 queues has
been changed a few times. The system might indicate that there are not enough resources to increase
queue size, even though there are enough free packet buffers. Removing and reapplying the policy map
on the interfaces solves this problem.
Use the show pxf cpu queue summary command to see the number of packet buffers, packet buffers
being recycled, and free packet buffers.
Scaling Enhancements in Cisco IOS Release 12.3(7)XI2
Cisco IOS Release 12.3(7)XI2 provides increased limits with queue scaling and VC scaling.
Queue Scaling
At least two queues are allocated for every interface or subinterface for which separate queues are
created. The first queue is the default queue for normal traffic, and the second queue, known as the
system queue, is used for a small amount of router-generated traffic that bypasses the normal drop
mechanisms. For 32,000 VCs, this setup would require the allocation of a minimum of 64,000 queues.
While Cisco IOS Release 12.3(7)XI1 adds support for up to 128,000 queues, a more effective use of
these limited resources is realized by having the subinterfaces on a given main interface share the single
system queue of the main interface.
OL-2226-23
Scaling Enhancements in Cisco IOS Release 12.3(7)XI2
Cisco 10000 Series Router Software Configuration Guide
2-7
Advertisement
Table of Contents
Troubleshooting
