Michael986
asked on
Switch / Network setup
I have a question about network design for a relatively small Windows network.
There are 4 servers – Data/DNS, Print/AD/DNS/DHCP, E-mail (Exchange) and Accounts (SQL), plus 5 * 24/48 port network switches and about 70 workstations.
All users require access to the Data, Print and E-mail servers, but only the (10 user) Accounts department use the Accounts server.
The current setup is that all servers are connected to switch 1, and LAGs connect switch 1 to switches 2,3,4 and 5 (ie everything connects to switch 1)
All the comms gear is connected to switch 2, and the workstations (and printers) are connected to switches 3,4 and 5.
With the aim being maximum network speed (or alternatively, least potential for bottlenecks), rather than redundancy, would this be considered to be the best setup?
Is having all the servers plugged into the same switch a good idea?
Should I, for example, have an ‘Accounts’ switch which would have the Accounts server and all accounts workstations plugged into it?
Or should I go one step further and split the 4 servers, putting one on each switch.
And what about the DNS servers – should they be on different switches
Any other suggestions?
There are 4 servers – Data/DNS, Print/AD/DNS/DHCP, E-mail (Exchange) and Accounts (SQL), plus 5 * 24/48 port network switches and about 70 workstations.
All users require access to the Data, Print and E-mail servers, but only the (10 user) Accounts department use the Accounts server.
The current setup is that all servers are connected to switch 1, and LAGs connect switch 1 to switches 2,3,4 and 5 (ie everything connects to switch 1)
All the comms gear is connected to switch 2, and the workstations (and printers) are connected to switches 3,4 and 5.
With the aim being maximum network speed (or alternatively, least potential for bottlenecks), rather than redundancy, would this be considered to be the best setup?
Is having all the servers plugged into the same switch a good idea?
Should I, for example, have an ‘Accounts’ switch which would have the Accounts server and all accounts workstations plugged into it?
Or should I go one step further and split the 4 servers, putting one on each switch.
And what about the DNS servers – should they be on different switches
Any other suggestions?
fgasimzade
It does not matter where you plug your devices as long as you have enough bandwidth between the switches. I suggest to have gigabit links between switches
All servers in one switch should be fine. if you have replication/clustering you chould put servers on different switches, since if the one fails. the other will take over.
you should like always have a "core switch" where all the other switches are plugged into so thats right what you did and just "spread" the network from there. should be fine.
Regards Aamodt
you should like always have a "core switch" where all the other switches are plugged into so thats right what you did and just "spread" the network from there. should be fine.
Regards Aamodt
Please increase number of port in LAG to increase bandwidth.
Please use two ports in LAG so total four ports will be used to have connectivity between two switches.
Please use two ports in LAG so total four ports will be used to have connectivity between two switches.
ASKER
Could anyone expand a little on the 'why' behind the recommendations.
For example, why should you always have a 'core switch' - in my example, what's the argument against having the accounts PCs on the same switch as the accounts server, as that's where most of their traffic will be.
Also, the fact that we can set up LAGs to link the switches suggests that there is potentially a need to have more than just a single 1GB link between the switches. Which suggests that there should be some method behind the planning, rather than just 'plug anything in anywhere'.
For example, why should you always have a 'core switch' - in my example, what's the argument against having the accounts PCs on the same switch as the accounts server, as that's where most of their traffic will be.
Also, the fact that we can set up LAGs to link the switches suggests that there is potentially a need to have more than just a single 1GB link between the switches. Which suggests that there should be some method behind the planning, rather than just 'plug anything in anywhere'.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
The link between the switches is where your biggest time should go as this is always the bottleneck in these cases. consider the max no of ports you can spare for the link/trunk and use teaming/LACP to get as much bandwidth as you can.
Is there a way of measuring whether an existing LAG is causing bottleneck and needs expanding. Currently I've got two LAGs of 4 ports - maybe I'd be better off having one of 6 ports and one of 2 - how would I find out if this is the case?
There are three realistic ways to asses the links:
in general a 4 port LAG is pretty good unless you have particularly heavy traffic flow over the link.
Use the switch's diagnostics to assess the amount of traffic flowing (if the switch has this facility)
If the switch has bandwidth/traffic monitors you can save your self a lot of hassle by just looking how much traffic is flowing.
Monitor the traffic using wireshark or another network monitor to see how much bandwidth is used
More complicated but certainly a valid option.
Make an educated guess by assessing what traffic is flowing between systems on some paper
assess how many machines are sharing the link and what bandwidth is available and you can estimate what bandwidth is available to each system. factor in any server traffic or high bandwidth users and you'll have an idea if the link will struggle or not.in general a 4 port LAG is pretty good unless you have particularly heavy traffic flow over the link.