|
I'm not quite sure I understand the importance of having multiple interfaces
on the LVS.
Worse comes to worst... the IP layer goes down on an interface. In that
case, the LVS stops receiving updates. It then thinks that the master must
have gone down and it should become the new master... so it becomes the new
master. When its IP layer comes back up it'll see that there is another
master already up and those two will elect who should become the new master.
The original master will have been a master for a lot longer and the machine
which just came back up will go back into slave mode.
If you're worried about that machine coming back up with the wrong
information in its tables... What's the big deal? It won't be the master
anymore and by the time something fails over to it, it will have the most up
to date information.
Have this protocol burp out the entire table every BLAH interval and just
send out changes every once in a while -- perhaps every time something new
is added to the table.
----- Original Message -----
From: "Joseph Mack" <mack@xxxxxxxxxxx>
To: "Wayne" <wayne@xxxxxxxxxxxxxxx>
Cc: "Ratz" <ratz@xxxxxx>; "Wensong Zhang" <wensong@xxxxxxxxxxxx>;
<lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Sent: Monday, August 07, 2000 1:34 PM
Subject: Re: ideas about kernel masq table syncing ...
> On Mon, 7 Aug 2000, Wayne wrote:
>
> > Agree that if IP layer hosed, we will be in trouble. Intel making a PCI
> > ethernet card has four 10/100 port on one PCI card.
>
> the only quad ethernet cards I've seen are about $1000/quad. Quite
> expensive. Are the quad Intels any cheaper?
>
> Joe
>
> --
> Joseph Mack mack@xxxxxxxxxxx
>
>
>
>
|
