cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
226
Views
0
Helpful
3
Replies

Server connection network design

annowong
Level 1
Level 1

Hi folks,

 

Here is the situation, my company is using WS-C3650-48TS as the core switch. Currently it connect with 16 client switch (HP V1910), 2 uplink to firewall (HA) and 2 uplink (EtherChannel) with 3560G-24TS which connect to about 10 servers with about 20 ports.

So, here is the question, would it be better to connect all the servers to the core switch and cut out the 3560G-24TS server switch? Most of the server traffic are between server and client, not server to server.

 

What are the reasons for and what are the reasons against this arrangement? 

 

Thanks!

3 Replies 3

Seb Rupik
VIP Alumni
VIP Alumni

Hi there,
Typically core switches have limited switchport density and high cost per port. One of the reasons for having a n-Tier network topology is to use an access layer where swithport density and cost are more favourable. When uplinks for the access-layer exceed what is available on a core switch, you would then introduce an aggregation layer.

 

What you have implemented is a 2 tier topology which would be considered best practice.

Unless you are seeing contention on the etherchannel between the 3560G to the 3650 core there is no real need to move the server connections. Doing so would probably only gain a minuscule drop in latency.

 

cheers,
Seb.

Hi Seb,

 

Thanks for your information.

 

Here is some further info in my case. For the core switch, as mentioned, currently it only occupied 20 ports out of 48 ports while there is no further plan to expand client switch connection, so there are about 28 ports available for use. In addition, we are planning to implement HA in the core switch by stacking an additional WS-C3650-48TS. If we want to implement the same in server switch, there will be an additional cost for purchasing 3560G-24TS with stacking module.

 

So, I am thinking would that be necessary to keep the server switch? Or would that be something good in practice to keep the current setting.

 

Thanks!

Since you are adding another switch to your core to form a stack you are doubling your port density and removing one of the restrictions/ concerns with connecting end-user devices directly to it.

 

In this case I would totally collapse your topology and connect your servers directly to the core switch with their port-channel member links split between the two switches.

 

cheers,

Seb.

 

Review Cisco Networking products for a $25 gift card