cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
4117
Views
10
Helpful
3
Replies

Circuit Utilization - Bandwidth Upgrade - Best Practice

mlitka
Level 2
Level 2

Can any one point me in to some documentation regarding best practices for determining when to upgrade bandwidth on a circuit?

I realize this can vary due to the applications in use, etc.

I would like to use our average circuit utilization graphs as a starting point to make this decision.

Is there a rule of thumb, say 50% average utilization, that should be used to determine when more bandwidth is required?

3 Replies 3

Scarecrow44
Level 1
Level 1

Mlitka,

 

I have included the only official CISCO comment on your topic.  But have also included some comments to help you.

 

The problem with using 50% average utilization is that depending on the traffic pattern you may be setting a standard that would flood your circuit.  Example:  You have a regular user base - 9am-5pm M-F but a 24 hour network.  Are you capturing traffic 24 hours or during the prime busy hours?  Do you include Saturday and Sunday?  If you are - the majority of your traffic could be minimal.  Are you including failover?  Can you re-route traffic in case of circuit failure? 

 

My network has a blend of circuits types with and without resiliency.  For some I use a simple failover calculation.  For others, we use 70% - 95th percentile utilization for prime busy hour/busy day traffic. 

 

The following is an excerpt from the following URL:

https://www.cisco.com/c/en/us/products/collateral/routers/wan-automation-engine/white_paper_c11-728551.html?dtid=osscdc000283

 

"The simplest core capacity planning processes use passive measurements of core link utilization statistics and apply rules of thumb, such as upgrading links when they reach 50 percent average utilization, or some other general utilization target. The aim of such simple processes is to attempt to ensure that the core links are always significantly overprovisioned relative to the offered average load, on the assumption that this will ensure that they are also sufficiently overprovisioned relative to the peak load, that congestion will not occur, and hence that the SLA requirements will be met.

 


@mlitka wrote:

Can any one point me in to some documentation regarding best practices for determining when to upgrade bandwidth on a circuit?

 

I realize this can vary due to the applications in use, etc.

 

I would like to use our average circuit utilization graphs as a starting point to make this decision.

 

Is there a rule of thumb, say 50% average utilization, that should be used to determine when more

bandwidth is required?


 

 


 

 

Craig Budrodeen
Level 1
Level 1

The problem with upgrading circuit bandwidth to overcome high utilisation stats is that TCP windowing will probably just absorb any extra bandwidth you throw at it.

If you know the nature of the traffic on the link, a better approach to managing congestion may be QoS.

 

A typical QoS configuration on a link that mixes interactive traffic with batch file transfers and real time traffic, may be to categorize traffic inbound by marking the ToS bits, then applying bandwidth management outbound where the real time traffic goes through a priority queue. The interactive and file transfer categories have Weighted Random Early Detection (WRED) applied with a bandwidth percentage applied to each category. That way the person who is interacting with a web server (say) processing an order doesn't compete with someone downloading a file, and the person watching a video stream gets priority to reduced jitter (and the all too familiar "buffering" error message).

The bottom line is user experience. If your users are experiencing poblems then choose the strategy best suited to what they use the link for.

Resilient TCP error correction means that while a congested link will cause TCP retransmits, this is part and parcel of the internet and does not necessarily mean poor user experience. And is why TCP/IP and the internet is so successful.

If you can monitor the rate of TCP retransmits via a packet capture and Wireshark, anything constantly above (say) 50% you need to take measures to fix it.

Personally I find measuring TCP retransmits a far better metric to determine what the user is experiencing rather than looking at the bandwidth utilizations on a graph which are stateful polls and do not capture what is happening in between polls.

dperezoquendo
Level 1
Level 1

Hello,

 

I don't think there is a rule of thumb anymore. There are too many factors to consider these days. Perhaps the following documentations may help you determine a "rule of thumb" for your network:

 

Capacity and Performance Management: Best Practices White Paper:

https://www.cisco.com/c/en/us/support/docs/availability/high-availability/20769-performwp.html

 

Best Practices in Core Network Capacity Planning:

https://www.cisco.com/c/en/us/products/collateral/routers/wan-automation-engine/white_paper_c11-728551.html