Calculating Switching Capacity and traffic ratio

Unanswered Question
Jan 8th, 2008


Good Day,

Kindly, i want to ask two questions , the first question about calculating switching capacity, how can i calculate the switching capacity in the core switch based on the number of users and amount of traffic that these users generate?!... For example, in my company, i got 600 users who are accessing the server farm that is consists of 10 servers , What switching capacity shall i consider for the core switch? ... Is there any rule shall i follow to calculate the core's switch capacity?

Another question about traffic ratio, if i have 48 desktop users equipped with 10/100/1000 NIC and all are connected to an edge switch 48-port 10/100/1000, how much bandwidth shall i give to the uplinks?




I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 4 (1 ratings)
Edison Ortiz Wed, 01/09/2008 - 19:29

> For example, in my company, i got 600 users who are accessing the server farm that is consists of 10 servers ,

> What switching capacity shall i consider for the core switch?

1) You won't be able to connect 600 users to a single switch, so you will have a bottleneck somewhere.

If you purchase an access switch (let's say a 3560) for your users (10/100/1000) and the users connect at 1Gbps, you will have a bottleneck on the uplinks if you have more than 2 users on that switch sending data at full speed. Therefore, you won't get much out of having users connecting at 1Gbps unless the servers are on the same switch.

2) Then comes the switch question, the answer it depends on the switch. Catalyst 3560 and 3750 support wire speed transmission within their switchport (on the uplinks you may be facing bottleneck on the aggregated bandwidth). 4500 and 6500 switches are modular and the answer will depend on the modules you've purchased for those switches.




obscured28 Thu, 01/10/2008 - 11:55

Hi Edison,

I appreciate your valuable answer regarding my question about switching capacity but here is the full picture:

I have total of 600 users in my company are connected to many access switches (of course, not to a single switch). These users are distributed based on department role .. let's say 70 users in sales Dpt., 90 users in the marketing Dpt. .... etc. The access switches are uplinked to a 4510R catalyst classic core switch and this switch provide a switching capacity of 136 Gbps using supervisor engine 6-E. Assuming that (just assuming) all the users are operating in 10/100/1000 (1 Gbps) and all are sending at the same time (of course, its nearly impossible having 600 users sending at the same time but just for assumption). In your opinion, do you think that there'll be a bottleneck at the core? ... oh by the way, the edge switches are cisco 2960.

In other words, shall i calculate it this way or am i wrong:

600 users x 1Gbps= 600Gbps which is >greater than 136 Gbps. (Bottleneck)

The second question is a general question and regardless of specific switch model, If i have a 10/100/1000 edge switch and all users operating at this speed, how much shall i give the uplinks? 10Gig uplink do you think?

I appreciate your fast response.

Thanks in advance


Edison Ortiz Thu, 01/10/2008 - 12:54

The bottleneck will be at the access switch.

If you have 1 user at the access switch sending at full speed (1Gbps), it will max out the uplink to the Core. If you were to have every single end-user connected at 100Mbps, the uplink will max'd out with 10 concurrent user connections.

The 4510R will also have bottlenecks at the module level itself (if you were to connect devices on the same module). While the switch forwarding capacity is rated at 136Gbps, each module has a 6Gbps connection to the backplane.




bobc Wed, 01/09/2008 - 20:24

Hi Obscured28,

The first thing you should do is to perform a baseline analysis for your existing traffic with tools like MRTG and PRTG. After that, you would have a better idea on the actual traffic behavior of your company by identifying average user and server traffic volume, bottleneck, peak traffic period...etc

In addition to traffic capacity, you should also consider the level of resilience on your network. Using two 3750/6500 switches in a stack/VS configuration would provide additional capacity as well as protection from single point of failure for your core/server farm.



jwdoherty Fri, 01/11/2008 - 18:16

You didn't note how many ports and of what bandwidth your 10 servers had, but assuming they each only had 1 gig port, then worst case you would only need 10 Gbps of bandwidth for your core. This also assumes your desktops don't normally communicate with each other. (Don't forget to allow room for growth.)

One rough rule of thumb, for servers, their busy ratios are offen in the range of 4:1 to 8:1, so your 10 servers, again with single gig connections, might only need about 2 Gbps of bandwidth.

One rough rule of thumb, for workstations, their busy ratios are offen in the range of 24:1 to 48:1, so desktop users with gig connection might work well with only 1 or 2 gig of uplink bandwidth.

As one of the other posters noted, actual traffic analysis would allow you to be more precise. More importantly, is selecting equipment where you can easily add bandwidth since often one finds hotspots. For instance, CADD workstations often use much more bandwidth then routine business desktops. So, 48 CADD users with gig connections might work better with a 10 gig uplink.


This Discussion