Unsolved

This post is more than 5 years old

2 Posts

2160

March 15th, 2006 16:00

New Network Layout using 2700 series switches

I am revamping our current network layout.  We will be doubling our existing network to about 75 nodes.  Basically I have centralized all the connections to a single server room.  There are approx 60 workstations (all with static IP), 12 network printers (using print servers on each), and 4 servers.  Three of the servers are function dependant and do not require much bandwidth.  However one server will be used for domain control, i-net gateway, terminal services, and centralized backup.  I am also toying around with housing each user's profile and/or my docs, etc on it.  Therefore I plan to use two NICs on it.  I am new to LAG, but it sounds like this is a good solution.  I currently own 2 2724 switches (both unmanaged).
 
Here is my idea....
Purchase 2 more 2724s, for a total of 4.  These will remain unmanaged and provide connections to each of the workstations and printers.
Then, purchase either a 2716 or 2724 to use as a aggregation point and configure it to managed mode.  Connect a single CAT6 line from each of the 4 unmanaged switches to this managed one.  Then connect each of the servers to it via a sing CAT6.  The high use server will have two NICs, so I will two CAT6 cables and modify the switch for LAG.
Finally, I will use a rack mounted APC SMART UPS 750RM to provide clean backup power to all the switches.
 
I am not a networking expert, just a software engineer.  Does this sound like a logical, sound approach?  Any comments are welcome!

March 15th, 2006 18:00

I'm not a network engineer either :-) - also a SW engineer but anyway here are my opinions:

  • You might do some network analysis to see if you really will be pushing 2Gig/sec worth of traffic to the server?  It's quite a bit of traffic and if you are using it internally servicing only 75 hosts you might not really need a LAG.  Simpler is better in any network :-).
  • But if your server really is that busy, then a LAG will provide you some addition in BW capacity but understand that the LAG distribution scheme is such that for any given source/destination pair the same port of a LAG is always used.  For example if you are connecting just a single client to a single server through a LAG that consists of two ports, the session connecting those two systems will always use the same port of those two ports in the LAG so even if you exceed the BW limit of the single port the overflow will not be sent through the other port.  That's not how a LAG works.  If on the other hand you have multiple clients connected to the server then the LAG distribution scheme will use the combination of source/destination MAC addresses to determine which of the ports in the LAG to use for any given source/destination.  So multiple clients may be distributed across the ports in the LAG.  The distribution would be "balance" only if there are randomness in the MACs of the source/destination.  I put a link on a paper below that describe this algorithm in detail.

So I would say that use a LAG if you think you need it.  It will provide better performance then if you have nothing but you should understand how it operate so that you have realistic expectation and can properly manage the network.

http://www.dell.com/downloads/global/power/ps2q05-20040286-Holmes-OE.pdf

Cuong.
No Events found!

Top