Nortel 2424 Load Balancing Switches Vs W2K NLB


We are running a high demanding web/application server farm.  Our web servers run on ports 80TCP and some of our applications run on UDP ports.

We are currently load balancing our servers in the farm using Nortel 2424 Load Balancing Switches.  Since the servers are W2K Advanced Servers, I thought of NLB as a more cost effective alternative.

I could not find any documentations or newsgroup discussions about the performance difference between a hardware load balancer such as the 2424, and the Windows built in NLB.  When looking at it from a performance point of view, which one is better?  What's the difference really?  What happens under the hood?  I understand that the 2424 uses Virtual IPs that distributes to the servers' IPs.  NLB on the other hand is set by having all farm members share the same VIP.  Does that make it slower?  High Performance is very critical for us, but I'd still like to know if NLB is an option and why.

Who is Participating?
I wear a lot of hats...

"The solutions and answers provided on Experts Exchange have been extremely helpful to me over the last few years. I wear a lot of hats - Developer, Database Administrator, Help Desk, etc., so I know a lot of things but not a lot about one thing. Experts Exchange gives me answers from people who do know a lot about one thing, in a easy to use platform." -Todd S.

Nortel is by far better than NLB:

1. Nortel is hardware based and is done for high performance. It can balance the traffic according to difficult and complex algotrighms without any performance loss, which is almost wirespeed.

2. Windows is a software, cannot compete in performance.

In case performance and also stability is a factor, stay with Nortel. Load Balancers cluster is usually a single point of failure on the way to the servers - you won't want it to die on you.
jmelikaAuthor Commented:
Thanks for the quick post.

When you say Nortel is for high performance, why?  Could you explain to me how that traffic is distributed at wirespeed versus Windows and why its way is slow?

I am looking for understand the complexity of how they both work rather than just which one is better.

I am not exactly familiar with functional abilities of NLB, but performance like here are some points:

1. Nortel runs on a small OS, dedicated and optimized specifically for these exact functions (Its either unix like system (some Linux distro adapted and optimized) or a VX Works which is known for its performance abilites.

2. NLB on the other hand runs on a non-optimized and not dedicated only for load balancing Os - Windows!

3. Nortel has dedicated processor for this specific functionality, while Windows runs on an Intel CPU which is not optimised as well...

Ultimate Tool Kit for Technology Solution Provider

Broken down into practical pointers and step-by-step instructions, the IT Service Excellence Tool Kit delivers expert advice for technology solution providers. Get your free copy now.

I looked again at your post and it seems to me that you want to understand how load balancing algorithms work, is it true?
There are few of them, I can explain it to you if you want.
jmelikaAuthor Commented:
yeah please!  That would be very helpful.

Well there are these several that are most used:

First we need to understand the setup:

- 2 or more servers are connected to the LB.
- 1 or more connections from users to the servers.

LB has a Virtual IP to which all connections are addressed.
It gets the packets and looks inside their contents and according to the contents it distributes them between the servers according to the LB algorithms that I am about to explain.

I will explain you here the LB on Layer 4 (transport) and Layer 7 (application).

Layer 4 switching (Load Balancing) is done in a following matter.

1. Round Robin: Connections are distributed between all servers one by one in a cyclic manner.
2. Hash or Persistent connections: all connections from the same user (IP address) are redirected to the same user.
3. Least Load : Load Balancer maintains the table of live connections and send each new connection to a least loaded server.
4. Least amount of connections
5. Least amount of users (each user can have more than 1 connection).

All these 3 mentioned above are LB for the same service type, meaning all service have the same functionality (Mirrors like).

Another is to distrbute connections according to service type or port number: Lets say we have an HTTP and SMTP service running behind the LB - all the connections will arrive to the same Virtual IP. So LB will accept the connection and will redirect it according to the service type or a port number to the correct server.

Also there are many other types of load balancing on different networking layers: from Ethernet to Applications layer. Different algorythms are suitable more or less to different types of traffic/applications and setups.

This is a liitle about load balancing.


Experts Exchange Solution brought to you by

Your issues matter to us.

Facing a tech roadblock? Get the help and guidance you need from experienced professionals who care. Ask your question anytime, anywhere, with no hassle.

Start your 7-day free trial
jmelikaAuthor Commented:
Well this is definitely a whole lot! Thanks! :)

Here is a thought that I have and please correct me if I'm wrong.  I'm really interested in learning as much as possible about this.

Windows NLB:
Each server in the LB cluster has its own dedicated IP.  They all share a completely separate IP (separate from their individual) and share listen to that specific port that's configured to be load balanced.  Let's take Round Robin for this example.  Server1 will OWN the MAC Address bound to the VIP.  This routes the next connection to that VIP over to Server1.  Now Server1 is doing the dew with the connection, Server2 on the list will OWN the VIP next, and so on.

Hardware LB:
The device is configured based on IP addresses in a VIP group and not by reassigning the MAC address to a new server each time.  Meaning, it does a form of routing for the traffic going to the VIP over to one of the IP's in the LB group.

That being said, of course after you verify that it's correct, does it make HW LB MUCH faster than Windows NLB?  If what I said is true, I'd be safe to assume so.  Correct?

Thanks again!
Again, sorry but I can't verify your statement regarding the NLB.

Regarding the HW solution you are correct - there is no need to change the MACs for the ports coz they are real. The physical ports (interfaces) have to be assigned to the "farm" with VIP as well, so VIP will listen on all assigned physical ports, therefore the MAC can be virtual as well.

Regarding the functionality when there is a fully utilised bandwidth and its full of many types of traffic - HW solution will take software windows solution big time - its dedicated, that simple. All large Data Centers use the HW solutions. Software is cheaper, but weaker and less reliable.

I hope this helps.
jmelikaAuthor Commented:
Thanks!  You definitely answered my questions.

It's more than this solution.Get answers and train to solve all your tech problems - anytime, anywhere.Try it for free Edge Out The Competitionfor your dream job with proven skills and certifications.Get started today Stand Outas the employee with proven skills.Start learning today for free Move Your Career Forwardwith certification training in the latest technologies.Start your trial today

From novice to tech pro — start learning today.

Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.