Switch/Router - Server Connection

Posted on 2002-05-02
Last Modified: 2008-03-10
In our datacenter we have approximatley 100 servers.  We plan to implement dual NIC connections for each server (teaming) back to the switch/router (Enterasys Networks). We are considering putting each server in it's own subnet, in the event a NIC goes bad and starts a broadcast storm and for performance.  Is there any reason we should not try this approach? Does anyone have this in place today?
Question by:ccherry13

Accepted Solution

jgarr earned 100 total points
ID: 6987777
I would think that putting each server in its own subnet will waste a lot of CPU on the router, not to mention IP addreses. Cisco switches have the ability to suppress broadcast storms at a port level for exactly this situation. Does Enterasys support this feature?

It goes back to the adage, "switch where you can, route where you must". Make normal sized subnets /24,etc and let the switch protect you from faulty NICS.
LVL 79

Expert Comment

ID: 6987794
I agree with jgarr. Putting a router in front of every server (virtually, not physically) defeats the purpose of putting them on a switch. Unless the switch has a layer3 capability on the backplane to act as the router. You might get better performance using Gigabit NIC's on your servers or fault-tolerant nic teaming. MSoft servers can load-balance over them quit nicely now.

A broadcast storm from a bad nick is a very rare occurance, and a good switch will suppress them.

Expert Comment

ID: 6988401
From your post, I gathered that you already have a router virtually in front of the servers.  I could be wrong, but I will base my comments on that assumption.  I disagree with jgarr's comments about CPU load to the extent that unless you have a very overwhelmed router already, 99 extra routes isn't going to affect it that much.  If you have a good IP scheme going already, IP addresses should be in very good supply and the "waste" should be relatively minimal.  

I will say though, that adding this much addressing for a single server farm is a major administrative overhead of managing those addresses - especially since servers are manually addressed.  I also agree with lrmoore's statement that broadcast storms are rare, so it seems like a lot of work for little gain.  At my full-time job, we have over 100 servers in a single subnet, and broadcast storms have not proven to be an issue.  Jgarr was also correct that a good switch will suppress problems here anyway.
Guide to Performance: Optimization & Monitoring

Nowadays, monitoring is a mixture of tools, systems, and codes—making it a very complex process. And with this complexity, comes variables for failure. Get DZone’s new Guide to Performance to learn how to proactively find these variables and solve them before a disruption occurs.

LVL 17

Expert Comment

ID: 7055987
This is only speculation since I don't know your network, however nic teaming leads to one of two things, redundancy or increasing bandwidth. If you need fault tolerance because your susceptible to down time then this is a good thing. Before you do it for bandwidth reasons, look at your servers first and analyze their network performance. In most cases the network is not the bottleneck, the server itself is. Before I would put any machine on it's own subnet I would see what gain I would get from doing it. First of all you will have the administrative overhead of reconfiguring each machine. Next, you will need to configure a router to be able to route all subnets, more administrative overhead. I see no gain from this at present. I would look more closely at server usage and see if there are any problems with network performance on individual machines and also do a network study to see if you are coming close to any network limitations within your datacenter.
LVL 79

Expert Comment

ID: 7997491
No comment has been added lately, so it's time to clean up this TA.
I will leave a recommendation in the Cleanup topic area for this question:

I recommend: points to jgarr

If you would like to keep this question open for more expert input, this cleanup effort will get it closer to the top of the list where it will get more visibility for the experts.

if there is any objection or other expert commentary to this recommendation then please post in here within 7 days.
If you feel that your question was not properly addressed, or that none of the comments received were appropriate answers, please post a request in Community support (with a link to this page) to refund your points.


EE Cleanup Volunteer

Expert Comment

ID: 8054166
per recommendation

Community Support Moderator @Experts Exchange

Featured Post

Resolve Critical IT Incidents Fast

If your data, services or processes become compromised, your organization can suffer damage in just minutes and how fast you communicate during a major IT incident is everything. Learn how to immediately identify incidents & best practices to resolve them quickly and effectively.

Question has a verified solution.

If you are experiencing a similar issue, please ask a related question

Hello , This is a short article on how would you go about enabling traceoptions on a Juniper router . Traceoptions are similar to Cisco debug commands but these traceoptions are implemented in Juniper networks router . The following demonstr…
Creating an OSPF network that automatically (dynamically) reroutes network traffic over other connections to prevent network downtime.
After creating this article (, I decided to make a video (no audio) to show you how to configure the routers and run some trace routes and pings between the 7 sites…
After creating this article (, I decided to make a video (no audio) to show you how to configure the routers and run some trace routes and pings between the 7 sites…

733 members asked questions and received personalized solutions in the past 7 days.

Join the community of 500,000 technology professionals and ask your questions.

Join & Ask a Question