help explaining the answer pls

Answered Question
Jun 14th, 2008

Q : What is the recommendation when planning for bandwidth requirements at the access to distribution layer links?

A: The oversubscription ratio should be no higher than 20:1

Can anyone pls explain in detail or give an exemple.. pls... Thanks

I have this problem too.
0 votes
Correct Answer by scottmac about 8 years 6 months ago

Given that the network is multi-tiered (access, distribution, and core), the reference is the ratio of potential ingress data to the amount of bandwidth available going towards the next tier up (access-->distribution).

If you had a single 24 port switch at the access layer, and every link was operating at 100Mbps, then you'd have a ration of 23:1 (23 ports to the client computers to 1 port going up to the distribution layer).

According to the question, that would exceed the recommended over-subscription (20:1).

To remedy the situation in this case, you could hard-set the client ports to 10Mbps (230 Mbps in, 100 Meg out, a 2.3:1 ratio), OR you could take one of the client ports, convert it to a Fast Etherchannel, which would give you a ratio of 22 client ports in, going to a logical port of 200Mbps 2200:200, or 11:1 ... either of which brings the ingress:egress ratio to under 20:1.

The idea behind oversubscription is that with exceptionally rare cases, only a few of the ingress ports will be active at any given time. So, rather than providing 100% guaranteed bandwidth to all traffic on every client port, the egress port bandwidth is "shared" amongst the ingress ports.

The amount of oversubscription should be adjusted on a network-by-network basis, because the type of host activity and the bandwidth needs are usually very different from one application to another.

A web-based service would generally require less bandwidth less often than a native database application (resident on the client) where records are routinely circulating between the client and the server(s).

That's why, as part of any network update/upgrade it's usually a "Good Thing" to do bandwidth studies for each segment to evaluate what the typical application flows look like, and what they will likely require when the {host count | application update} is implemented.

All those numbers are compared to the budget and available hardware, adjusted for user needs / priorities, and an oversubscription ratio number is arrived at. 20:1 is a decent number, I s'pose, as a rule of thumb, but it should be understood that the number can and probably should vary according to the needs (tempered by the budget) of the network and the users.

Good Luck

Scott

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (1 ratings)
Loading.
Correct Answer
scottmac Sat, 06/14/2008 - 08:22

Given that the network is multi-tiered (access, distribution, and core), the reference is the ratio of potential ingress data to the amount of bandwidth available going towards the next tier up (access-->distribution).

If you had a single 24 port switch at the access layer, and every link was operating at 100Mbps, then you'd have a ration of 23:1 (23 ports to the client computers to 1 port going up to the distribution layer).

According to the question, that would exceed the recommended over-subscription (20:1).

To remedy the situation in this case, you could hard-set the client ports to 10Mbps (230 Mbps in, 100 Meg out, a 2.3:1 ratio), OR you could take one of the client ports, convert it to a Fast Etherchannel, which would give you a ratio of 22 client ports in, going to a logical port of 200Mbps 2200:200, or 11:1 ... either of which brings the ingress:egress ratio to under 20:1.

The idea behind oversubscription is that with exceptionally rare cases, only a few of the ingress ports will be active at any given time. So, rather than providing 100% guaranteed bandwidth to all traffic on every client port, the egress port bandwidth is "shared" amongst the ingress ports.

The amount of oversubscription should be adjusted on a network-by-network basis, because the type of host activity and the bandwidth needs are usually very different from one application to another.

A web-based service would generally require less bandwidth less often than a native database application (resident on the client) where records are routinely circulating between the client and the server(s).

That's why, as part of any network update/upgrade it's usually a "Good Thing" to do bandwidth studies for each segment to evaluate what the typical application flows look like, and what they will likely require when the {host count | application update} is implemented.

All those numbers are compared to the budget and available hardware, adjusted for user needs / priorities, and an oversubscription ratio number is arrived at. 20:1 is a decent number, I s'pose, as a rule of thumb, but it should be understood that the number can and probably should vary according to the needs (tempered by the budget) of the network and the users.

Good Luck

Scott

Actions

This Discussion