Last Comments .....

1) re Mathematics -- I think this is valuable but there are two difficulties
   a) If Math not precise, then it is confusing. I see this at some places
   b) One needs to set up the Math but also have discussion of operational
      issues like "how best to set up Network of 22 servers"
      I think this (operational issues) is weak area

2) I think you need to add three significant areas
   a) Performance (most important). Thus in chapter 4, the key ideas are not
   really protocol but how and why and with what results. 
   What are good topologies
   How does "matching" process translate in good performance so that messages are
   not sent to unnecessary servers and are routed to clients in some roughly
   optimal fashion
   b) Fault Tolerance as an operational issue (a) and b) guide how
   one should set up a network of servers)
   c) Implementation Issues. How you structured code. Where is data buffered etc.

3) section 2.7 needs more examples. It is very hard to read

4) In chapter 8 describe 22 server set up. It looks odd to me. Why would you
   set up such a network like a snake (few connections per server). I think
   a more clustered set up looks beeter

5) describe a model for GES -- JMS has a very clear publish-subscribe model
   State somewhere GES model

6) Discuss persistence a little. Is "system persistence" different from 
   "archiving". It is in JMS

7) For each chapter state the ESSENTIAL idea and somewhere collect all your ideas
   Perhaps at the end of thesis where you can list key concepts
   	Publish/Subscribe
   	Performance
   	Accuracy (merging etc.)
   	Routing
   	Persistence etc
   	
   	and your approach to them.
   	
8) State that clients and servers can be on same physical machine (sec 2.2)

9) Discuss dynamic topologies. it seems to me that source/destination matches
   actually tell you best server network and in an optimal system, you would
   dynamically create servers

10) Note parallel computing analogy
    servers are Network
    clients are comute nodes
    
11) Section 2.1 "The sub vector" .... confuses me

12) Sec. 2.3. I think you should discuss "implicit" destinations as the nature
   of implicitness determines problem

13) is Sec. 2.5 d) really correct -- statement seems bizarre to me

14) Discuss where matching decision made. In "best" commercial JMS,
    refinement in subtopics made at receiving client which is silly

15) Chapter 8 -- 
    motivate example
	Why in sec 8.2 isn't bandwidth important
	Isn't product of bytes size * message important measure
	Why isn't variation of latency versus hop length very interesting
	Wouln't you design servers so hop length small?
	I would add results hop length 1 4 10(done) 18 for selected cases
	Add 1 server results (same as hop length 1?)
	Your results are given to too many digits. Three significant figs is good choice