LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: PCC patch for the VS 0.8

To: Peter Ke{e <peter.kese@xxxxxxxxxxxxxx>
Subject: Re: PCC patch for the VS 0.8
Cc: "Shi, Eddie E" <eshi@xxxxxxxxxxxxxxxxx>, "'Peter Kese'" <peter.kese@xxxxxx>, virtual server mailing list <linux-virtualserver@xxxxxxxxxxxx>
From: Wensong Zhang <wensong@xxxxxxxxxxxx>
Date: Wed, 28 Apr 1999 22:43:19 +0800
Hi Peter,

Peter Ke{e wrote:

> Hi!
>
> The patch is not incorporated into VS 0.8 but is a separate patch based
> on VS 0.8.
>
> There are some minor things I would like to work out on that patch, but
> I am currently very busy with porting the Virtual Server to 2.2. The PCC
> scheduling algorithm will be 'officially' released probably only in
> the 2.2 kernel series.
>

I have never meaned that I want to nix the Persistent-Client Connection
feature in the virtual server patch. Although it has some dynamic load
imbalance, for some applications the two connections from the same
client may be related and must be assigned to the same server, the PCC
is a solution and can be kept as an option like the LocalNode feature.
However, there are some minor problems that we need to discuss and fix.
For example, an SSL key has a life span of 100 seconds after the end of
a connection which uses it, after which it expires, so there is no advantage
to preserving the persistent connection beyond that point. And, whether
additional masquerading entries are created needs discussion too.

I want to make that users can set a port (a virtual service) as "sticky"
(persistent), and set the expire time of the "sticky" port by using of
ippfvsadm. I also want to make that the request dispatching techniques
for a port can be changed on-the-fly by using of ippfvsadm too, I mean
changing the request dispatching techniques without rebuilding kernel.

By the way, if the HTTP cookie data, SSL connection data and so on
are shared by all the servers, can I assign the two connections (related
like above) to the different servers?


>
> Hopefully this will happen soon. I am currently testing the VS on the
> 2.2 kernel and it works fine. For the moment I have skipped the tunneling
> forwarding option but have implemented direct routing instead.
>

Is there any problem in porting the tunneling feature to kernel 2.2?

>
> Before I publish it, I would like to test it on a SMP machine, remove
> my debugging calls and prepare a proper kernel patch out of it.
>
> That shouldn't take me more than two or three days.
>
> Cheers,
>                         Peter
>

Good luck,

Wensong





<Prev in Thread] Current Thread [Next in Thread>