|
First, you limit the servers to be Linux only. Although the original
poster did not mention that, but we could assume they are Linux.
2nd, DSL line is not as reliable as the T1/T3 lines, if traffic
coming from a line suddenly dead, the whole site will be all dead. This
will make the reliability totally depends on that one line, reduce the
reliability to 1/10.
At 12:50 PM 7/6/00 -0700, Ian S. McLeod wrote:
>I believe it should be possible to use IPIP tunneling in this situation.
>
>Use a single load balancer connected to one of the DSL lines.
>
>Use the remaining DSL lines for real servers.
>
>Point all incoming traffic at the load balancer (as usual). The balancer
>will select a real server and forward the request packets to it through
>the IP tunnel. (You could even use your backend NFS network for this to
>avoid the extra traffic over the DSL lines.).
>
>All response packets will then go out through the 9 DSL lines connected to
>the real servers. Assuming this is mostly web traffic with a
>response/request bandwith ratio of about 10:1 you should be able to
>saturate the outgoing bandwith on your DSL lines (if the load is well
>balanced :).
>
>Can anyone see a reason why this wouldn't work?
>
>-Ian
>
>On Thu, 6 Jul 2000, Jon Oringer wrote:
>
> > I understand the LVS application for 1 internet connection (say 15 MBits)
> > and 10 servers - each of which will have an equal load...
> >
> > But -- My situation is a bit different.. I have 10 separate 1.5Mbit DSL
> > Lines.. (a total of 15Mbits)... and 10 servers.. Each box serves web pages
> > only (via apache).. Right now I planned to use only round robin DNS for each
> > of the 10 servers to balance the load.. there is a sep. disk servers
> > avail via. NFS. Each box looks exactly the same from the web (they all
> > share the same web
> > directory via. NFS).
> >
> > The examples for LVS describes one internet line, and multiple servers..
> >
> > Can LVS be used for multiple internet lines, and multiple servers?
> > Or should I just stick with round robin DNS?
> >
> > Thanks!
> > -Jon
> >
> >
> >
> >
>
>
>
|
