Mailing List Archive

Luca Pescatore wrote:
> this is my first installation of mod_backhand and so 'm very newbie.
> my actual configuration is with two server, web1 (IP 10.0.0.62) and web2
> (IP 10.0.0.63).
> httpd.conf 10.0.0.62 :
> UnixSocketDir /usr/local/apache/backhand
> MulticastStats 10.0.0.255:4445
> AcceptStats 10.0.0.0/24
>
> httpd.conf 10.0.0.63 :
> UnixSocketDir /usr/local/apache/backhand
> MulticastStats 10.0.0.255:4445
> AcceptStats 10.0.0.0/24

Looks good. As note, you should make sure that your netmask/brodacast
are a /24 before setting it up this way. Many OSs default to broadcast
10.255.255.255 and netmask 255.0.0.0 for th 10/8 network. Just for
awareness ;)

> http://servername/backhand/ :
> Entry Hostname Age Address Total Mem Avail Mem # ready servers/
> # total servers ~ms/req [#req] Arriba # CPUs Load/HWM CPU Idle
> 0 web1 0 10.0.0.62:80 263913472 235345088 6/6 0 [0] 913927 1 0.000000/1
> 1.000000
> 1 web2 0 10.0.0.63:80 264003584 250578112 0/0 0 [0] 913950 1 0.340000/1
> 0.960000

Looks like it is working... ;)

> ok, now my question, where i need to set-up www.domain.com ? if .62 goes
> down, .63 get requests instead of .62 ? How can i do that ?

Hmm, this is a whole other ball game.

First, a simple description of what mod_backhand is: a resource
allocator...
Second, a description of what mod_backhand isn't: a high availability
tool...

That should answer your question right there. So, mod_backhand won't
direct to machines that have failed, but making sure that requests do
not originate to a failed machine is out of the scope of mod_backhand.
There are tools that can do that:

Linux VirtualServer, Cisco LocalDirector, F5's BIG/ip, Alteon switches,
Foundry switch, ExtremeNetworks switches, the list goes on. But,
mod_backhand plays games on the application level. Games need to be
played on the network level in order to assure high availability.

If you want cheap, use Linux Virtual Server (free).

If you don't want to dedicate a box to this, then write a program that
will bring up a virtual interface on a live machine in response to a
peer going down. If you are working on only two machines, or if you
only want peer-peer failover then this problem is straight forward (use
the free product called 'fake'). N-way reliable and controlled failover
(on an Ethernet/IP level) is a HARD problem... We are working on it at
www.cnds.jhu.edu ;)

So, you may ask "If I use Linux VS or buy a BIG/ip or Extreme box, why
would I use mod_bakhand?" Because the algorithms in mod_backhand are
smarter and COMPLETELY customizable... Network level load balancers
just are good enough .. yet.

--
Theo Schlossnagle
33131B65/2047/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7