Calculating oversubscription on access layer

Unanswered Question

So, the situation is that we have about 240 users (max, no further expansion to this figure) on each floor of a building, spanning across 10 floors. As I understood from the basic calculation, for every 240 users @ 100/1000, I need to deploy 5(48 port stacked switches within each IDF on each floor) (5x48) will share 2x1GigE uplinks(etherchannel) terminated into two different cores with a redundant 2x1G uplink (passive).


Someone tells me that this will result in an oversubscription ratio of 120:1 in the best case scenario when we fully utilize both uplinks in an active/active setup. Is this ratio unacceptable at access layer?


Should I consider dual 10GigE uplinks from the access layer to core, to start with, irrespective?


I currently am working with 1G uplinks and don't see any major hassles. My applications though utilise 3D and CAD drawings but I still feel that 2-4Gig uplink would be more than sufficient for me as I can link aggregate further upto 8Gigs using Cisco Cat 3750E.


Should we go for 10Gig uplink straightway?

Your thoughts!!

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 4.5 (2 ratings)
Loading.
lgijssel Wed, 10/10/2007 - 03:10
User Badges:
  • Red, 2250 points or more

The scenario where all users attempt to tranmit&recieve at full speed simultaneously is purely theoretical, this will never happen in real environments. I have not yet seen a 1G link that was continuously congested by user traffic.

Instead of only looking at the max througput, you should also consider the average bandwidth consumption per user. This is mostly a somewhat more modest figure.

Another factor determining the actual amount of overbooking is the aggregate bandwidth that is available from the servers. Example: if there was just one server running at 1Gb/s, an uplink of 2G will never become saturated.


When you feel that 2Gig is adequate for the time to come, you need not go to 10G from the start. This technology (as all) will become cheaper over time.


regards,

Leo

Joseph W. Doherty Wed, 10/10/2007 - 20:27
User Badges:
  • Super Bronze, 10000 points or more

You really need to look at your traffic endpoints to determine what the possible bottlenecks are. If most of your users are only going to a couple of servers, then it's likely the bottleneck will be the server links, not you access uplinks.


With that in mind, a 10 gig link works better than channeled gig links. First, you don't have the problem of multiple flows using the same saturated gig link of a bundle while others are not being used. Second, you don't need to concern yourself that the correct channel hash method is being used.


Although 10 gig ports are more expensive than gig ports, if you factor in the possible need for additional cable runs and the cost of multiple gig modules, 10 gig might becomes less expensive sooner then you might expect.


You note you don't have any problems with gig uplinks today. Assuming your moving users from 10/100 to 10/100/1000 and they will be using 1000, unlikely performance will be any worse, but possible it won't improve much either.


PS:

Deja-vu - reminds me of moving from 10 to 100 Mbps for users and 100 to gig for uplinks

Actions

This Discussion