|
Hi,
Sorry for the late reply.
Dronamraju Rajesh wrote:
>
> Hi there,
>
> I have configured lvs - nat using pulse, nanny, ipvsadm, etc. I am using
> kernel 2.2.14 on red hat 6.2
Poor you :) Consider takeing a more stable and ready for productionary
release of the lvs-patch. Grab a fresh 2.2.18 kernel, the lvs-patch and
you're much more happy on how things will work for you.
> Now my doubt is as follows:-
> I get continous http requests to my set of servers in production. Now I
> cache every request that comes to each of the servers. If another request
> comes there-after I match the cache entry and deliver the response for that
> request in order to avoid the request to go through the database. Please
> note that I get nearly 1.5 to 2 million requests per day for each server.
Wow! What's the site? What kind of hardware to you use to be able to handle
such a lot of requests a day using VS-NAT and even do database access?
> Now if I configure lvs NAT with persistance, the requests coming from
> one IP will be going to the same server until the time parameter for
> persistance configured is reached.
Exactly. But we can savely assume that not everybody is coming from the
same IP your following statement is not true:
> Hence if I implement lvs with or with out persistance, there will be
> heavy load imbalance and lvs will work only for fail over concept but not
Of course the load imbalance is high. I've seen load imbalance myself but
this was always in short time view. In a long perspective you will see that
things flatten down. Install mrtg or rrdtool and do some graphs. And then it
also highly depends on the architectural setup of your service. Is persistency
really needed for all access to your 'webpage'?
> for load balancing. Please correct me if Iam wrong. So, how do I implement
> this concept in lvs as Iam completely fascinated by the way it works.
Maybe you like to have a look at the new scheduler lblc/r.
Best regards,
Roberto Nibali, ratz
--
mailto: `echo NrOatSz@xxxxxxxxx | sed 's/[NOSPAM]//g'`
|
