Re: [jgroups-users] Multicast UDP without higher level protocols?

On 12/02/16 00:37, Questions/problems related to using JGroups wrote:
> Greetings form Oregon!

Greetings from Switzerland!

> I have a question about the higher level protocols when it comes to UDP
> multicast:
>
> I was working on some existing code tracking down some instances of what
> appeared to be possibly a split brain.  Their configuration was only the
> lowest level multicast: (Sorry about the old style, just wanted to
> replicate exactly what they had)
>
> UDP(mcast_addr=228.9.7.9;mcast_port=45599;
> ucast_recv_buf_size=128K;ucast_send_buf_size=128K;mcast_recv_buf_size=128K;mcast_send_buf_size=128K)

This is as if you only used a MulticastSocket for sending and receiving 
of messages; no group membership or retransmission.

> So I created a test where a JChannel was created connecting to
> “Docker-JGroups” cluster and periodically “ping” with
> channel.send(Message)  I’d then watch the channel view and the results
> of who replied for each hosts.
>
> I’d then fire this up in a batch of docker containers (# ranging from 3
> to 20)
>
> I’d then tweak the host iptables to either drop or reject the multicast
> packets coming through the docker “network”  I’d see the individual
> programs cease to find each other and be running in their own instance.
> That’s good
>
> Then I’d delete the iptables rule and I’d see them all join up again
> into a single network cluster? (By single cluster, I mean, all the hosts
> are responding to the JChannel.send() again)
>
> I thought things like FD, MERGE and pbcast.GMS were required for all of
> that to work properly?  In fact the only thing I can tell is that when I
> have only UDP protocol setup, that each node thinks it’s the cluster
> coordinator, and failure detection seems to be a little faster with all
> the other protocols combined, but I can’t tell any difference otherwise?
>
> Obviously I’m missing something basic – would you mind explaining why
> the behavior is so similar despite the fact that there’s only one
> protocol configured with this particular configuration?

With UDP only, the members were not really in a cluster; they each fired 
messages to the multicast address and everyone received them. However, 
this did not
- provide a cluster view (channel.getView() should be null) and notify 
the app when members joined or left
- retransmit dropped packets
- prevent duplicate packets
- provide ordering

etc.

I've seen stacks like UDP:UNICAST3 which means that multicasts are 
unreliable and unicasts are reliable, the use case was audio/video 
transmission where dropped packets are not a problem.

-- 
Bela Ban, JGroups lead (http://www.jgroups.org)


------------------------------------------------------------------------------
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140