1. Home
  2. Cisco Systems
  3. Newbie: REQ: Peer Review: Proper use of a Catalyst 4948

Newbie: REQ: Peer Review: Proper use of a Catalyst 4948

Eight years ago I put together a web farm for a client, but haven't networked since. Somehow that qualifies me to move a web farm from a managed hosting company to a co-location facility this month. :-O

I've put together a network map

formatting link
) and I would really appreciate a peer review on the network design and whether or not the planned Cisco Catalyst 4948 usage is proper.

This network primary focus is serving over 5 billion (with a B) http GET requests per month (6,000 GET requests per second at top peak times). 99.9% of the traffic is HTML and JavaScript files (AVG 3K in size), no email, no desktop support, no ftp, no images, no streaming, you get the idea.

Questions:

1a. This first draft runs all the traffic through a Cisco Catalyst 4948. Is this flawed idea?

1b. My reading of the Cisco docs suggest this will not negatively impact the performance. Do you agree?

  1. I plan on purchasing a second 4948, but I don't understand if and/or how I can setup a hot failsafe. Is my only choice a cold spare?

  1. Do you see any other newbie errors? If so, please suggest how I should go about correcting them.

Thank you in advanced.

Reply to
Newbie
Loading thread data ...

My opinion is that you should put in the second 4948 ASAP and configure them for HSRP for your external (outside Load Balancer) and internal (inside Load Balancer) networks. I would connect one load balancer to one switch with as much bandwidth as possible (2 gig ether channel or something if possible), and the second load balancer to the second switch. I would also dual home the firewalls to both switches, and this way you could lose a switch, fiber trunks, a load balancer, etc, and still be in service. I support hundreds of environments (production, lab, and test), and this would be a common architecture, although we use much larger switches......

Additionally, and assuming your servers have two NICs, you would home each server to each switch, so full redundancy is built in.

Reply to
Trendkill

Trendkill,

Thank you for your insightful suggestions. I think I've incorporated all your ideas. Please have a look and let me know if I overlooked anything:

formatting link
This design leaves me with one question....

With all the servers are on two networks 192. and 10., which I understand this is a good thing for redundancy, I'm confused how the web farm and app server's code will find another server in the event a switch goes down.

Example: the web farm calls the database by using it's address

192.168.20.10. The 192. switch goes down, 10. switch is operational. Don't the calls to the 192.168.20.10 fail because it's not using the 10.10.20.10 address? If it doesn't fail, why not? Is there something I need to do in the switch configuaration. I'm sitll getting my head around HSRP, maybe it's handled here?

Thank you for your any additional help you can provide.

Reply to
Newbie

Well, and perhaps my idea needs some re-engineering. Technically, both networks should run on both switches. A virtual IP address will be assigned as a gateway (usually .1), and each switch would have an IP in each network (usually .2 and .3). This way, if one switch fails, the other will still be able to route for both networks.

However, I seemed to have ignored the fact that you need each box to connect to two different networks, thus eliminating your ability to connect one to one switch, and one to another, all in the same network. If the first switch fails, you are correct that your servers will only be able to talk out of the second interface, which needs to be in a separate network. Therefore, I would recommend connecting your load balanced servers across the switches instead. If the LB is balancing web server 1 and web server 2, connect both NICs from server

1 to switch 1, with 1 connection in each network. Connect server 2's connections to switch 2. That way if a server died, the other will still pick up the traffic, and if a complete switch died, you would still have half your servers up while you tend to the outage. Let me know if this makes more sense. Feel free to email me directly via here and we can chat via email or phone.
Reply to
Trendkill

Unless of course you have 4 NICs in each server. Then do one connection to each switch in each network, then you will have full redundancy even if a switch fails.

Reply to
Trendkill

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.