infinispan usage

[Paul Hethmon]

On 9/7/11 2:45 PM, "Chad La Joie" <lajoie at itumi.biz> wrote:


>On Wed, Sep 7, 2011 at 14:34, Paul Hethmon
><paul.hethmon at clareitysecurity.com> wrote:
>> Yea, I've had to learn a lot about multicast and there is more that I
>> don't know. But it looked like the UDP multicast was the recommended
>> network topology, though with a small cluster, it probably doesn't
>>matter
>> a lot. The one nice thing about multicast, at least for my environment,
>>is
>> that I don't have to explicitly make the configuration know about each
>> node. Maybe the TCP option can do the same, but I haven't looked at it
>> closely.
>
>Well, I think that's why most of the examples show using the
>multi-cast.  The reason I decided not to use it for HA-Shib was, first
>and foremost, because most networks at the time didn't seem to like
>routing multi-cast message across subnets/vlans.  I don't know if
>thats really changed or not.  In addition, at the time, there wasn't
>any mechanism for restricting who could join the cluster.  If you knew
>the multicast address and port, you were in.  That meant that if your
>network did allow cross-subnet/vlan multicasting then some one else
>sitting out on the network could join the cluster and get your state
>data (which might be sensitive).

Still all true. I've had to enable multicast on my office network to
develop. And there is still no security. So if you're crossing LAN
boundaries, or have things on your subnet that you don't control, it's not
a good option. In my case, I control the boundary router and will keep it
from passing multicast.


>
>> What I'm trying to achieve though is the easiest way to configure a
>> cluster. I'm running about 60 or 70 Shib clusters right now. Not all of
>> them are configured for session support and hence need clustering, but
>> enough of them. So I should be able to get away with only having to
>>change
>> the port configuration for each cluster to keep them separate. I can
>>tell
>> them to use 239.0.0.1 as the mcast_addr and then change ports. So
>>cluster
>> 1 gets port 10000, cluster 2 gets 10100, etc.
>
>Why are you running so many clusters?

That many distinct customers. Then add in the 400k to 450k distinct end
users and we are probably doing 1.5m authentications a day. I might have
to actually query my database and see what my aggregate numbers are, never
have done that.

Paul