0MQ: The Intelligent Transport Layer

[Roy Williams]

On 12/26/11 8:08 PM, Bob Denny wrote:
> Magic Question: What is wrong with
> http://www.ivoa.net/Documents/Notes/VOEventTransport/

Nothing wrong with vTCP/Dakota. I am happy to report that Skyalert 
supports it. However, I have only heard of one instance of an 
observatory actually making use of events distributed this way (NMSU), 
so I am also looking elsewhere.

There are other protocols. The users of the original GCN socket will ask 
why not that one? Or the fixed-format emails that MPC uses, they say 
what is wrong with that transport. Both the CRTS and LIGO Lab have 
invested in Jabber/pubsub. I like http as a protocol too. We use both 
XML and JSON for events. So a profusion....

I would say the most important thing is interoperability with VOEvent. 
All of the above protocols and formats are happy to carry something 
semantically representable as a VOEvent, and in many cases actual 
VOEvent syntax. In other words, VOEvent uptake is improved by providing 
what people want, and by translating what they give us to VOEvent. We 
cannot just cry about being in the same corner since 2006 and why is 
nobody coming!

As for scaling, I would say there are various avenues. Sheer bulk of 
message is one dimension, burst rate and average rate, also number of 
streams, number of selections on streams, number of users, size of 
linked data, compute requirements for annotation, required reliability, 
degree of replication and distribution, and several more. Lets us ask 
which of these dimensions are crucial to scale, and which not.

Roy

---
Caltech LIGO
roy at caltech.edu
626 395 3670