LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: Thank to all of you! And couple of questions.

To: Skliarouk Peter <skliaroukp@xxxxxxxxxxx>
Subject: Re: Thank to all of you! And couple of questions.
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Lars Marowsky-Bree <lmb@xxxxxxxxx>
Date: Fri, 3 Sep 1999 15:42:25 +0200
On 1999-09-03T13:05:09,
   Skliarouk Peter <skliaroukp@xxxxxxxxxxx> said:

> Does ipvsadm supports "smooth" deleting of an server from it's server's
> table: Not to redirect any more connections to the server, but NOT to
> break any existing connections (for server maintenance)?

Thats actually what happens with the 0.7 patch in 2.2.12-final, but it is not
intentional ;-)

Looking at the ChangeLog for the latest 0.8.1 however,

-   Added the dest server status checking.
    The server status is checked before forwording a packet. If the
    server is not available(down or put out of service), the packet
    will be dropped and the client will be notified immediately.

    The server status is also checked while generating a masq entry
    based on the masq template. If not available, the new entry won't
    be created.

What would be needed here is a way to differentiate between paragraph a and b.

Looking at the code, this _could_ be achieved by setting the weight of the
server to 0 for wrr scheduling and would require a special check for wlc
scheduling (even if I don't see the weight being used in the wlc algorithm at
all, but maybe I am blind).

> Is there a way to hook up (somehow) request-type based preference
> (e.g. all requests for static pages should go to one server and
>    all dynamic requests to several another servers)?
> (like eddie promised me to do (which I couldn't install)).

This can not be done using the LVS code, since it requires knowledge of the
data stream, which we can't easily get at. Easily done from user space, but
that has other properties to consider.

However, what could be implemented is something like Cisco's "flow switching".
The first packet (or in our case the first data packet) is switched by a user
level process, which then creates an additional masquerading entry / template
so all further packets get "flow switched" by the kernel.

This might be easier to implement using Netfilter in 2.3/2.4. Comments?

> Not directly linked: How I could export (via SNMP) server's
> CPU load/disk load ?

UCD SNMP.

> PS. I'm not subscribed to the mailing list - please answer me directly.

Please subscribe, we try to concentrate discussions about the project on the
mailing list, so everyone can benefit from them. To subscribe, send an e-mail
(body & subject are ignored) to lvs-users-subscribe@xxxxxxxxxxxxxxxxxxxxxxx

Thank you!

Sincerely,
    Lars Marowsky-Brée
        
--
Lars Marowsky-Brée
Network Management

teuto.net Netzdienste GmbH - DPN Verbund-Partner

----------------------------------------------------------------------
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
To unsubscribe, e-mail: lvs-users-unsubscribe@xxxxxxxxxxxxxxxxxxxxxx
For additional commands, e-mail: lvs-users-help@xxxxxxxxxxxxxxxxxxxxxx

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