Re: [load balancing] Alteon UDP load balancing issue

From: Henrik Lantz <henrik [izzat] onyx.nu>
Date: Thu Jun 26 2008 - 01:38:38 EDT

Hi Ed,

As far as I've been able to tell, the application does not send keepalives.
The delta time between the request and the response ranges in the 20-200
millisecond range at the moment (the application is still being tuned), so I
wouldn't expect the Alteon to have a problem maintaining state for that
short a time period. (I see a PIX firewall we have a bit deeper in the
network tearing down states after about 5 minutes or so.)

The interesting thing is that I can't see a state being set up in the Alteon
session table for the request at all, something which puzzles me. The
request obviously DOES hit the load balancer as it's coming out on the
inside with the RIP address as the destination IP. Hmh.

Thanks,

    Henrik

----- Original Message -----
From: "Ed Toro" <cojonesdetoro@gmail.com>
To: "Load Balancing Mailing List" <lb-l@vegan.net>
Sent: Thursday, June 26, 2008 4:34 AM
Subject: Re: [load balancing] Alteon UDP load balancing issue

> Could it be due to a UDP state timeout issue? TCP conversations have
> flags that tell you when someone is "done" taling (RST flag?). Devices
> managing UDP conversations need to maintain state information about
> on-going conversations. Of course, this means allocating memory that
> eventually has to be freed up so it must have a timeout so it can free
> up the memory when it seems that the conversation is finished. Does your
> application send keep-alives while it's searching?
>
> Henrik Lantz wrote:
>> Hi Bruno,
>>
>>
>> The application I am trying to load balance is a proprietary one, built
>> by one of our vendors. It is basically a catalogue service listening to
>> a UDP port, performing a quick database lookup and then sending data back
>> to the client. As UDP is a connectionless protocol, I guess you could
>> technically say that the server is setting up an outgoing connection; but
>> since it is in response to a client request directed to a VIP on the load
>> balancer, shouldn't the load balancer know to keep state? The response
>> connection is sent back to the same port that originated the request,
>> i.e. (seen between load balancer and server):
>>
>
> _______________________________________________
> lb-l mailing list
> 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

_______________________________________________
lb-l mailing list
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
Received on Thu Jun 26 01:39:00 2008

This archive was generated by hypermail 2.1.8 : Thu Jun 26 2008 - 01:39:04 EDT