Router and circuit size for headquarter - quantitative decision

Unanswered Question
May 31st, 2008
User Badges:

Folks, can you give me your thoughts on this. Based on the attached diagram and application usage (people on remote sites will use primarily a client/server application named MyAPP). Based on real data captured on a similar site in production right now, I included the peak rate and data transferred over one week period. I am trying to make a decision on what type of router and circuit based on quantitative information.

Basically I am considering the peak rate as an important factor when determining the circuit size. Does that make sense to you?

My concern is this:

If I have the information that the peak rate for the aforementioned application is is 4.0Mbps for a circuit in a given remote site, if I haved the total of 5 sites in hte future, that by itself tells me that I need a bigger circuit than 9Mbps in the headquarter side?

Can you please see the diagram and help me answer the following:


Is the 3825 in the headquarter the right size of router if I plan to add four

More sites? (each new site is expected to introduce data usage rate).

If I add four more sites in the future, do I need to upgrade the circuit in the headquarter to a bigger circuit?

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Giuseppe Larosa Sat, 05/31/2008 - 12:19
User Badges:
  • Super Silver, 17500 points or more
  • Hall of Fame,

    Founding Member

Hello Marlon,

I cannot see your attachment.

However, I suggest that you have to take in account the type of application that is used. If the MyAPP application is based on TCP you can use a primary link of 9 Mbps without big problems. TCP flows use the bandwidth that is available and adapt (through the windowing mechanism) to changing conditions. Of course lesser bandwidth means slower response but if the increase in delay is acceptable for the application and from the end user's prospective you achieve your goal. Dimensioning the network using peak values can be too expensive. You can also make use of QoS to provide better treatment to business related traffic (like MyAPP) then to Internet www traffic.

A C3825 is enough to do the job.

If you use any form of encryption this requires more resources.

hope to help


Joseph W. Doherty Sat, 05/31/2008 - 17:51
User Badges:
  • Super Bronze, 10000 points or more

Assuming you want MyAPP to avoid network queuing latency, you might try using the average utilization during the busiest couple hours during the work day, and then size the circuit's bandwidth such that the average utilization during peak times is somewhere between one third to two thirds of the circuit's bandwidth (or the circuit is 3x to 1.5x your peak hours' average utilization). If you only have FIFO queues at the bottlenecks, and you treat all traffic alike, move toward the 1/3 usage. If you have multiple FIFO queues, and you can distinguish interactive traffic from bulk traffic, try about 1/2. If you have WFQ and not too many bulks flows (that have less weight than the interactive), or multiple classes queues each with FQ (few platforms support this type of policy), then move toward 2/3 usage. (Also, as the circuit gets larger, you can usually also move toward the higher utilization rate.)

When sizing a circuit, often the traffic bandwidth is asymmetric, if it is, size for the busier direction.

Cisco recommends router sizing as:

2801 - 1xT1/E1 (i.e. about 4 Mbps)

2811 - 2xT1/E1 (i.e. about 8 Mbps)

2821 - 4xT1/E1 (i.e. about 16 Mbps)

2851 - 6xT1/E1 (i.e. about 24 Mbps)

(ref. Table 1 in:

3825 - half T3 (i.e. about 22.5 Mbps)

3845 - full T3 (i.e. about 45 Mbps)


But these recommendations, I believe, are very conservative since they allow for all possible services at full line rate. I found many of the above routers can usually deliver much more performance if not all services are used.

Attached is a Cisco router performance sheet seen often within these forums. I suggest you might size based on about 25% of the CEF Mbps rate.


This Discussion