|
On 2000-04-27T11:37:41,
Wayne <wayne@xxxxxxxxxxxxxxx> said:
> I have found that there is a problem for using LVS persistent modes
> to handle AOL customers during load balancing. We have about
> 30% customers are from AOL. However, AOL has each user
> goes out to fetch a page through several cache proxy servers.
> I could see that one user gets a page that has a few gif files is
> by getting the html code from cache-rg07.proxy.aol.com, the next
> gif file from cache-rg04.proxy.aol.com, the next gif file from
> cache-rg05.proxy.aol.com,... So the source IP address for even
> a single page is not consistent.
Yes. We had the same problem with the German Telekom proxy servers.
This was why I added the quick hack of "persistence netmask". Setting this to
--netmask 255.255.255.0 is most likely going to solve your problem.
For a documentation of what this does, look at the ipvsadm man page.
The _real_ solution would indeed be to keep the session data in a centralized
(maybe replicated) database. Maybe LDAP could be used for this, as it
replicates between its instances by itself and could be run on each node.
> The persistent mode needs to be able to load balancing based on
> the actual browser, somehow.
The other option would indeed be to allow LVS to schedule on the http request
itself. This is not going to be easy to add - Wensong is working on that
though I believe. Until then, you may want to investigate mod_backhand for
Apache or using an inverse application level proxy like Squid's http
accelerator feature.
Sincerely,
Lars Marowsky-Brée <lmb@xxxxxxx>
Development HA
--
Perfection is our goal, excellence will be tolerated. -- J. Yahl
|
