Forum Discussion

lcw1982_107422's avatar
lcw1982_107422
Icon for Nimbostratus rankNimbostratus
Feb 24, 2008

Load balancing across multiple clusters

Hi,

 

 

I am pretty new to load balancing technologies, so need some help here.

 

 

I am thinking of using LTM across two sites, situated about 500m apart. Since the distance is pretty short, especially with the use of fiber cables between the two sites, I think it might not be necessary to use the GTM. Part of my network design is attached, the distribution switch connects to other part of the server farm.

 

 

I have some doubts which i need to clarify:

 

 

1. The linkage between the two sites will span across the distribution switches. May I know if load balancing will work across the two sites? Or do I need to connect the load balancers from the two sites directly?

 

 

2. I am using load balancing between different applications - app, web and terminal server. Can i share the load balancers for these 3 clusters, or do i need a separate load-balancer for each application? Does configuring multiple pooling address this?

 

 

3. May I know what benefits do Layer 4 and layer 7 load balancing provide? And also for this configuration, what specifications do I need to look out for in the distribution switch? Layer 2? Layer 3?

4 Replies

  • It is possible to load-balance servers that are at different sites, however there are some caveats.

     

     

    Most likely, if you are not spanning L2 across the sites, the servers will be on different networks. If the servers do not have the LTM as their default gateway, you will have to SNAT all incoming client connections so that when the servers respond, they will send the traffic back to the LTM before going back to the client. This causes all traffic in server logs to appear as though it comes from the LTM, which may or may not be a problem. You can somewhat alleviate this for HTTP traffic by inserting an X-Forwarded-For header which can be logged through various mechanisms. For non-HTTP traffic though there aren't such options.

     

     

    Also you must consider that health checks to the servers will have to be performed over the link between the sites and this could add delay to detecting server failures and additional chattiness on those links.

     

     

    LTM does not limit how many applications you can put on the box, the only limits are throughput, CPU, etc.

     

     

    L4 and L7 lb (such as with iRules, etc) allow much more intelligence in directing traffic but it does not mitigate the fact that the L2/L3 routing has to work underneath it all.

     

     

    Hope that helps,

     

    Denny

     

     

  • Hi Denny,

     

     

    thanks for your explanation. actually I'm quite confused as to how I can configure to allow the servers across both sites to share the same virtual IP and be load-balanced.

     

     

    The traffic would effectively come in from one of the firewalls. If both the web server in one site is busy or down, would the traffic be diverted via the layer 2 switch, to the FC switch and to the load balancers in the other site?

     

     

    I'm assuming I have to configure both sites as the same subnet since I need to configure clustering on some other applications.
  • I think I misunderstood slightly. I was thinking of the LTM's being only at one site. If you have LTM's at both sites then that's really a case for using GTM to be able to lb both sites, because you are right, you can't really have the 2 sites share the same virtual IP.
  • May I know if its possible to configure both sites as a single subnet, and I perform my load-balancing without the GTM?

     

     

    Also, i notice the LTM has 2 10Gb fiber ports. Would it be possibel to connect the load balancers across the two sites via the fiber link ports? or should I link the two sites via the uplink switch as shon in the above figure?

     

     

    Thanks in advance for any help.