Re: how to achieve URL hash LB algorithm on ZXTM

Hi, We use the Zeus core as part of our Enterprise range of appliances
as such we have written a lot of complex traffic/Java and have not
experienced any CPU limitations even in very busy environments. Clearly
it's a programming language so it will depend on the quality of the code
etc but Ben is right, it seems like we see a small Hit for a basic
script but no measurable difference for a large script.

Greg
 

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf
Of Ben Mansell
Sent: 31 July 2009 16:43
To: Load Balancing Mailing List
Subject: Re: [load balancing] how to achieve URL hash LB algorithm on
ZXTM

Excellent news!

It's difficult to say, in general, what impact TrafficScript has on 
performance, as it all depends on just what you are doing with it.
Complex 
actions, such as performing big regular expression matches and data 
substitution, will use more CPU. The rule that you are using should have

not much impact, as it is doing relatively little work, for instance.

A key point though is that there is no "fast path" for HTTP processing 
that cannot be used when you start using TrafficScript, i.e. you won't 
take a big hit to performance when using small amounts of TrafficScript
vs 
using none.

Ben

On 31/07/2009 15:24, Eric Y. Zhang wrote:
> Thanks Ben
> it works like a charm!! TrafficScript is quite impressive. any idea
about how TrafficScript impacts the performance of ZXTM?
>
>
>
> On Thu, Jul 30, 2009 at 6:33 PM, Ben
Mansell<[email protected]<mailto:[email protected]>>  wrote:
> That is what it achieves. It does this by using a session persistence
mapping that picks a server to use based on the URL requested, and not
on the client.
>
>
> Ben
>
>
> On 30/07/2009 06:45, Eric Y. Zhang wrote:
> thanks. but that is not what I really want. the trafficscript divided
traffic based on session, I want something like this
>
>
www.abc.com/1.html<http://www.abc.com/1.html><http://www.abc.com/1.html>
on squid1 permanently
>
www.abc.com/2.html<http://www.abc.com/2.html><http://www.abc.com/2.html>
on squid2 permanently
>
>
> Eric
>
> On Wed, Jul 29, 2009 at 5:19 PM, Ben
Mansell<[email protected]<mailto:[email protected]><mailto:[email protected]<mailto:ben
@zeus.com>>>   wrote:
>
> On 29/07/2009 00:48, Eric Y. Zhang wrote:
> routing requests to different backend server in same pool.let's say, I
have 5 squid servers behind LB, if LB can do URL hash, then 1of squid
server only needs take 20% requests, that helps a lot.
>
> There are a few ways to achieve this. The first is to use the
'perceptive' load balancing algorithm. This takes several factors into
account when choosing a node to use, including the URL of a request.
However, it's only one of many factors, so it doesn't rigidly enforce
the URL->squid server mapping that you probably want.
>
> What you can do instead is to leave the LB algorithm unaltered, and
instead use a session persistence class to dictate where each request
goes.
>
> Create a 'Universal' session persistence class and assign it to your
pool of squid servers. This then lets a TrafficScript rule pick which
squid server to use. The TrafficScript command is:
>
> connection.setPersistenceKey( http.getRawURL() );
>
> ...and then requests with the same URL will use the same squid server.
However, this needs a bit of refinement. You are likely to have many
many different URLs going to your squid servers. Each different URL
creates a different session persistence mapping, and there are limits on
the number of these. To ensure that too many aren't used, we can 'hash'
the URL into a smaller set of sessions:
>
> # Hash the URL, then take the first byte, giving 256 different
sessions.
> # This also uses the website hostname, in case you are caching several
> # websites.
> $session = string.left( string.hashMD5( http.getHostHeader() .
>                                         http.getRawURL() ), 1 );
> connection.setPersistenceKey( $session );
>
> So, add this rule in, assign it to your virtual server and then your
squid traffic will be nicely divided up. This should improve the cache
hit rates.
>
>
> Ben
> _______________________________________________
> lb-l mailing list
>
[email protected]<mailto:[email protected]><mailto:[email protected]<mailto:lb-l@
vegan.net>>
>
> http://vegan.net/mailman/listinfo/lb-l
> Searchable Archive: http://vegan.net/lb/archive
> http://lbdigest.com Load Balancing Digest
> http://lbwiki.com Load Balancing Wiki
>
>
>
> --
> Life is hard
>
> _______________________________________________
> lb-l mailing list
> [email protected]<mailto:[email protected]>
> http://vegan.net/mailman/listinfo/lb-l
> Searchable Archive: http://vegan.net/lb/archive
> http://lbdigest.com Load Balancing Digest
> http://lbwiki.com Load Balancing Wiki
>
>
>
> --
> Life is hard

_______________________________________________
lb-l mailing list
[email protected]
http://vegan.net/mailman/listinfo/lb-l
Searchable Archive: http://vegan.net/lb/archive
http://lbdigest.com Load Balancing Digest
http://lbwiki.com Load Balancing Wiki
_______________________________________________
lb-l mailing list
[email protected]
http://vegan.net/mailman/listinfo/lb-l
Searchable Archive: http://vegan.net/lb/archive
http://lbdigest.com Load Balancing Digest
http://lbwiki.com Load Balancing Wiki