Forum Discussion

IanK_37848's avatar
IanK_37848
Icon for Nimbostratus rankNimbostratus
Nov 04, 2009

Comparing Results of the various Loadbalancing Methods

Guys

 

 

I am trying to optimise the load balancing method we are using with distributing requests to a http/squid VS.

 

 

Currently we are using Ratio(member) due to various different hardware profiles of the servers , however I want to test the difference a dynamic method would make to the traffic levels and EU latency.

 

 

Question:

 

 

Is there any specific stats I can be looking at apart from connections on the VS ?

 

 

is there any way to measure the latency using a irule or 3rd party application sitting on a EU PC ?

 

 

 

Hardware , a LTM6800 cluster running BIG-IP 9.4.6 Build 425

 

 

 

Any advise or process guidelines on changing a LB method on a VS that peaks at 2GB , greatly appreciated

 

 

Thanks

 

Ian

8 Replies

  • Ian: if you're using a squid farm and you're caching, I'd strongly suggest you evaluate the election hash algorithm.

     

    An explanation is here: http://devcentral.f5.com/weblogs/dctv/archive/2008/06/26/3399.aspx as well as a link to the full solution.

     

     

    -Matt
  • Hi Matt , i am now researching your recommendation heavily as it looks perfect. ( as most people i wasn't even aware of that option.

     

     

    Details of our deployment. , we basically offer our customers the proxy( filtering solution ) and they use our proxy to fetch content from the Internet, the primary function is web filtering , but to ensure contents is served as quick as possible is is most definitely cached on the squids.

     

     

    Now whilst I am researching this fully , what stats or app can you use when benchmarking how much of a performance improvement / degradation a potential modification has.

     

     

  • spark_86682's avatar
    spark_86682
    Historic F5 Account
    And of course v10.x also has the election hash load balancing built-in, so you probably won't need an iRule. iRules are fast, but can never be as fast as real compiled code.
  • Spark: amen, and thanks for the reminder. Ian, the LB method in v.10 that you're after is called CARP, and as spark notes it's right there in the gui, ready to rock.

     

     

    -Matt
  • Guys .. thanks for that .. i am however struggling to find documentation for this on ask.f5 , the only reference for (Cache Array Routing Protocol) is on the configuration manual .. and it only mentions it not in depth detail at all.

     

     

    -• CARP: Specifies that the system uses the Cache Array

     

    Routing Protocol (CARP) to obtain the hash result for the

     

    input to the algorithm.

     

     

    there is nothing in the release notes about it .... or am i missing it ?

     

     

     

    Thanks

     

    Ian
  • Er, I should have said that the *persistence* method for this is CARP, not the LB method.

     

     

    -Matt
  • vinceKahn_97184's avatar
    vinceKahn_97184
    Historic F5 Account
    I could not find anything in the docs on how to set this up. I've done what I think is correct but am surprised to see when I refresh a request it hits all three of my proxies.

     

     

    I've created a Hash Persitence profile with CARP enabled. I created the following iRule and applied it to the new Hash profile.

     

     

    when HTTP_REQUEST {

     

    persist hash [HTTP::uri]

     

    }

     

     

    I then added the new persistence profile to my virtual server. I tested hitting http://www.whatismyip.com/ With each refresh I see one of my three proxy IP's. With the older election hash irule I'd always see one of the three. Let me know what I'm missing?

     

     

    Thanks,

     

    Vince