Wells Dairy Uses NetMRI
"NetMRI is one of those products that, once you see it in action, you wonder how you can live without it. We saw a good ROI and that was all it took to make the decision to install it on our network."
—Jim Kirby, Network Engineer
Read Case StudyConnect With Us
Tech Tips
Take the Next Step:
Related Information
Network Analysis Tip: Cisco Buffer Tuning
Why is this important?
Unnoticed packet loss in networks can adversely affect the applications using the network and one potential source of packet loss is in the store-and-forward buffering used in Cisco routers. Cisco routers generally do an excellent job at handling network traffic, but sometimes a bit of buffer tuning will improve the process. In some rare cases, an IOS bug will never release buffers, eventually consuming all of memory. Watching buffer creation and the total number of buffers (or memory) in use will alert you to places in the network where buffer tuning may be helpful.
Cisco divides system buffers into multiple sizes, which are used to hold packets being process switched, routing updates, telnet and ssh, and other purposes:
| Buffer Type | Size (in bytes) |
|---|---|
| Small | 104 |
| Middle | 600 |
| Big | 1524 |
| VeryBig | 4520 |
| Large | 5024 |
| Huge | 18024 |
Each buffer type can have a number of permanent buffers, as well as minimum and maximum free buffers. The router will create new buffers as needed to handle the packet flows.
Manual determination
Buffer statistics are displayed using the command show buffers. In the command output below for Middle buffers, there are 25 permanent buffers, 10 minimum, and 150 maximum. There are currently 150 total buffers, so the maximum number has been allocated. The difference between the created and trims counts is 125 (25 permanent + 125 created = 150 total). There are currently 3 buffers in use (150 total -147 in free list = 3 in use). The misses count is the number of times that a buffer was needed and one had to be created. Of these, the failures count is the number of times that the buffer could not be created in time to handle the packet and the packets was dropped. There were no occasions when buffer creation failed (the no memory count).
Middle buffers, 600 bytes (total 150, permanent 25):
147 in free list (10 min, 150 max allowed)
61351931 hits, 137912 misses, 51605 trims, 51730 created
91652 failures (0 no memory)
Network traffic can occur in two types of bursts: slow bursts and fast bursts. In a slow burst, the router creates new buffers as needed and can keep up with the burst. This causes the misses counter to increment. A fast burst does not allow enough time for the router to create buffers, so the packet is dropped, incrementing both the misses and failures counters. This is where hidden packet loss may occur in a network. Obviously, performing this check using manual methods on all the routers in a large network would not be done very often. Automated methods are required.
Automatic determination
Automated systems such as NetMRI monitor buffer misses, failures, and no memory counters on a daily basis and identify routers where these counts are excessive. It is also useful to know what time of day the buffer counters changed so that you can determine the type of traffic involved and the application that is likely to be affected. NetMRI creates an issue when a router is experiencing excessive buffer misses or failures, or when there are any no memory counts.
Corrective Action
In the above example, the maximum number of buffers exists in the pool, so it would be reasonable to increase the max count by a small percentage, say 20%, or 30 more buffers. The number of failures is .15% of the total number of buffer hits, so increasing the number of permanent buffers to 50 would be reasonable, given the buffer size of 600 bytes. The following commands would perform this change:
buffers middle permanent 50
buffers middle min-free 20
buffers middle max-free 180
Further reference:
Buffer Tuning for all Cisco Routers: http://www.cisco.com/warp/public/63/buffertuning.html