From wbell@CS.Cornell.EDU Wed Nov 28 21:22:16 2001
Received: from postoffice.mail.cornell.edu (postoffice.mail.cornell.edu [132.236.56.7])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fAT2MFT11463
	for <egs@cs.cornell.edu>; Wed, 28 Nov 2001 21:22:15 -0500 (EST)
Received: from [192.168.1.100] (syr-66-24-16-64.twcny.rr.com [66.24.16.64])
	by postoffice.mail.cornell.edu (8.9.3/8.9.3) with ESMTP id VAA28482
	for <egs@cs.cornell.edu>; Wed, 28 Nov 2001 21:22:14 -0500 (EST)
Subject: 615 PAPER #65
From: Walter Bell <wbell@CS.Cornell.EDU>
To: egs@CS.Cornell.EDU
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
X-Mailer: Evolution/0.99.1+cvs.2001.11.07.16.47 (Preview Release)
Date: 28 Nov 2001 21:21:52 -0500
Message-Id: <1007000535.2006.21.camel@brute>
Mime-Version: 1.0

65) Herald: Achieving a Global Event Notification Service


This paper discusses the desirable axes on which to build a global
event notification service, where the scalability requirements force
the ability for it to be deployed on the global Internet. They are
concerned with fine grained message / event delivery for interactive
activities such as instant messaging.

They've taken the approach that like the Internet itself, any global
event service must view the Internet as a set of mutually untrusting
federations of machines, owned by separate entities. They also insist
that the scale of node must be variable in that single nodes of one
federation might be equivalent to thousands of nodes in another
federation. This view really pushes the view that the goal is the
unite several large federation services into one coherent service,
versus starting from the ground up. With this view, it would be only
necessary to implement gateways from current massive event services
into the herald system in order to interoperate.

Since the paper is light on details and is more a treatise on design
issues in this space, it's only reasonable to inspect their decisions
in the design space. My major problem was the disconnection of naming
from this entire picture-- they seem to view naming as a completely
isolated problem and provide a weak infrastructure for helping a
naming service which would be necessary for herald. The fundamental
naming problem involves one of distribution of information which is a
portion of the problem they wish to solve.

They seem to be taking a good view with respect to simplicity-- many
of the other event services are very complicated especially with
respect to guarantees; I have to believe that any successful global
system must be simple with simple guarantees. The Internet has proven
that time and time again. It's not clear about their view on soft
state and event guarantees; it seems as if a more expressive construct
might be required for more diverse applications. I find it very hard
to believe that many applications would work reasonably in face of
rendezvous points disappearing/forgetting their subscription
information without notification. I think they've hit some points very
well, but that time will show them they haven't quite captured what is
necessary for scalability as well as a useful programming paradigm.



From viran@csl.cornell.edu Wed Nov 28 23:02:52 2001
Received: from naismith.csl.cornell.edu (naismith.csl.cornell.edu [132.236.71.13])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fAT42qT28707
	for <egs@cs.cornell.edu>; Wed, 28 Nov 2001 23:02:52 -0500 (EST)
Received: from localhost (viran@localhost)
	by naismith.csl.cornell.edu (8.9.3/8.9.2) with ESMTP id XAA23237
	for <egs@cs.cornell.edu>; Wed, 28 Nov 2001 23:02:47 -0500 (EST)
	(envelope-from viran@naismith.csl.cornell.edu)
X-Authentication-Warning: naismith.csl.cornell.edu: viran owned process doing -bs
Date: Wed, 28 Nov 2001 23:02:47 -0500 (EST)
From: "Virantha N. Ekanayake" <viran@csl.cornell.edu>
To: egs@CS.Cornell.EDU
Subject: 615 Paper 65
Message-ID: <Pine.BSF.4.10.10111282300470.23230-100000@naismith.csl.cornell.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

Slides at:

<a href="http://www.csl.cornell.edu/~viran/herald.pdf>
www.csl.cornell.edu/~viran/herald.pdf
</a>
<a href="http://www.csl.cornell.edu/~viran/herald.ps>
www.csl.cornell.edu/~viran/herald.ps
</a>

From eyh5@ee.cornell.edu Wed Nov 28 23:30:29 2001
Received: from memphis.ece.cornell.edu (memphis.ece.cornell.edu [128.84.81.8])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fAT4UST03488
	for <egs@cs.cornell.edu>; Wed, 28 Nov 2001 23:30:28 -0500 (EST)
Received: from photon.ece.cornell.edu (photon.ece.cornell.edu [128.84.81.138])
	by memphis.ece.cornell.edu (8.11.6/8.11.2) with ESMTP id fAT4RbM06559
	for <egs@cs.cornell.edu>; Wed, 28 Nov 2001 23:27:37 -0500
Date: Wed, 28 Nov 2001 23:27:40 -0500 (EST)
From: Edward Hua <eyh5@ece.cornell.edu>
X-X-Sender:  <eyh5@photon.ece.cornell.edu>
To: <egs@CS.Cornell.EDU>
Subject: 615 Paper # 65
Message-ID: <Pine.LNX.4.33.0111282326320.29056-100000@photon.ece.cornell.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

Herald: Achieving a Global Event Notification Service

Luis Felipe Cabrera, Michael B. Jones, Marvin Theimer

	
	This paper proposes Herald, a distributed system designed for 
event notification systems. Its targeted clients are those who subscribe 
to a certain service for access of information offered by that service. An 
example of an event-notification system is the MS Instant Messaging. 
Herald stands apart from other event-notification systems in that it is 
scalable to millions of users, a task previously achievable only in a 
centralized system. In fact, Herald is intended to expore the scalability 
issues involved in building a truly global event notification system with 
decentralized characteristics.

	The basic assumption in Herald is that any event notification 
system can be decomposed into a highly-scalable base layer that has 
relatively simple semantics and multiple higher-level layers. For this 
base model, Herald encompasses several functions, including hegerogeneous 
federation of machines, scalability, resilience, self-administration, etc. 
In the meantime, it avoids the inclusion of some higher-level functions 
such as naming, filtering, in-order delivery, and complex subscription 
queries.

	One of the core concerns for the researchers developing Herald is 
that it must be able to withstand network adversity with great tolerance. 
That is, Herald is presumed to operate in an adverse environment where 
nodes are mutually suspicious of each other's behavior, node failures are 
common, and the distributed state information is partial and incomplete. 
Herald attempts to overcome these adversities via means of service 
redundancy. The built-in redundancy is deployed to allow greater 
scalability and enhanced fault tlerance. Also, a time contract is imposed 
on maintaining the content of the service. This soft-state approach seeks 
to strike a balance between recycling precious disk space and 
accommodating the service. 

	Although Herald is a distributed system, it does, according to the 
designs of its researchers, include an administrative Rendezvous Point 
that performs the function of notifying all interested parties changes 
occuring in the Rendezvous Points that house subscribed services.  I do 
not understand why the Rendezvous Points themselves cannot communicate 
with each other and with service subscribers/publishers/creators about 
changes in the status of the Rendezvous Points. Employing an 
administrative Rendezvous Point may not be an efficient approach, and may 
very well become a bottleneck problem and a single point of failure.

	


From ramasv@CS.Cornell.EDU Thu Nov 29 01:12:53 2001
Received: from exchange.cs.cornell.edu (exchange.cs.cornell.edu [128.84.97.8])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fAT6CrT20460
	for <egs@popsrv.cs.cornell.edu>; Thu, 29 Nov 2001 01:12:53 -0500 (EST)
X-MimeOLE: Produced By Microsoft Exchange V6.0.4712.0
content-class: urn:content-classes:message
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Subject: cs615 PAPER 65
Date: Thu, 29 Nov 2001 01:12:44 -0500
Message-ID: <706871B20764CD449DB0E8E3D81C4D4301E7F29A@opus.cs.cornell.edu>
X-MS-Has-Attach: 
X-MS-TNEF-Correlator: 
Thread-Topic: cs615 PAPER 65
Thread-Index: AcF4nNvneNw95w3tSLSA6PjvQLfWlQ==
From: "Venu Ramasubramanian" <ramasv@CS.Cornell.EDU>
To: "Emin Gun Sirer" <egs@CS.Cornell.EDU>
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from quoted-printable to 8bit by sundial.cs.cornell.edu id fAT6CrT20460

Herald: Achieving a Global Event Notification Service

	This paper describes the desing goals for herald that aims to
create a internet scale event notification service.  Their proncipal
design feature is a rendezvous point that serves as the connection
between multiple subscribers and publishers.  Their world consists of
several rendezvous points that could be created and destroyed by clients
that act as subscribers or publishers .  In turn for reasons of
scalability and resilience each rendezvous point is replicated among
several servers.  One of the important design goal is to self organize
this replication feature and provide scalable interaction among the
replicated copies.

	The authors of this paper seems to favour the use of several
thousand small overlay networks to achieve scalability.  In their design
each rendezvous point would consists of replicated servers at
geographically distinct locations forming an overlay network while the
system itself could consist of several thoudand rendezvous points.  The
impact of this architecture on the internet could however be drastic,
significantly impairing the generally good functionality of the
internet.  However, the idea of self-organizing the replication process
based on needs of load distribution and fault tolerance appears to be
quite useful.

 

From c.tavoularis@utoronto.ca Thu Nov 29 01:21:09 2001
Received: from bureau6.utcc.utoronto.ca (bureau6.utcc.utoronto.ca [128.100.132.16])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fAT6L8T21988
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 01:21:08 -0500 (EST)
Received: from webmail2.ns.utoronto.ca ([128.100.132.25] EHLO webmail2.ns.utoronto.ca ident: IDENT-NOT-QUERIED [port 54544]) by bureau6.utcc.utoronto.ca with ESMTP id <239222-27436>; Thu, 29 Nov 2001 01:20:58 -0500
Received: by webmail2.ns.utoronto.ca id <24413-11979>; Thu, 29 Nov 2001 01:20:43 -0500
To: egs@CS.Cornell.EDU
Subject: 615 PAPER 65
Message-ID: <1007014839.3c05d3b73b48e@webmail.utoronto.ca>
Date: Thu, 29 Nov 2001 01:20:39 -0500 (EST)
From: c.tavoularis@utoronto.ca
MIME-Version: 1.0
Content-Type: 	text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7BIT
User-Agent: IMP/PHP IMAP webmail program 2.2.3

Herald is global event notification for distributed systems that can be 
employed by applications including instant messaging. It uses a federated 
approach, as opposed to centralized system, such that the defining task is 
interaction between peers. 

Herald is a simple system inspired by the Internet to allow scalability in 
terms of the number of users, the number publishable events, and the rate of 
event arrival. The gist is that publishers advertise data to a rendezvous 
point, and subscribers subscribe to the rendezvous points they are interested 
in. Rendezvous points are located on servers, and can change locations or even 
exist on multiple servers. Herald is in charge of creating rendezvous points, 
providing replication, and sending subscribers event notifications. Replication 
involves having copies of rendezvous points on more than one server. This 
provides fault-tolerance and load balancing, although all copies of data may 
not be complete. Information will expire after a time contract and publishers 
must explicitly refresh data.

Herald uses an overlay network to forward information and unicast or reliable 
multicast can accomplish event notification. Redundant messaging by replicated 
rendezvous points must be avoided. This not dealt with in this paper, but I 
think it could be done via some kind of agreement between replications such 
that they are each responsible for a unique set of subscribers. Of course, this 
information would have to be maintained and it will reduce the efficiency of 
replication in the event of a network partition.

Herald claims to support of heterogeneous federation, dynamic location of 
users, self-administration, timeliness and support for disconnections and 
partitions. The authors do not explain how they handle contradicting or 
incomplete information at different servers. Although correctness is not an 
explicit goal, it is possible that a subscriber could miss events over time and 
therefore Herald is only useful when events are not critical. Also, it would be 
useful for subscribers to be notified when they have been partitioned from the 
rest of the network. I think this paper is a work in progress and requires more 
detail.

From papadp@ece.cornell.edu Thu Nov 29 03:35:10 2001
Received: from memphis.ece.cornell.edu (memphis.ece.cornell.edu [128.84.81.8])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fAT8ZAT13198
	for <egs@CS.Cornell.EDU>; Thu, 29 Nov 2001 03:35:10 -0500 (EST)
Received: from kiki.ece.cornell.edu (kiki.ece.cornell.edu [128.84.83.13])
	by memphis.ece.cornell.edu (8.11.6/8.11.2) with ESMTP id fAT8WHM10131;
	Thu, 29 Nov 2001 03:32:17 -0500
Date: Thu, 29 Nov 2001 03:38:25 -0500 (EST)
From: "Panagiotis (Panos) Papadimitratos" <papadp@ece.cornell.edu>
To: Emin Gun Sirer <egs@CS.Cornell.EDU>
cc: Panagiotis Papadimitratos <papadp@ece.cornell.edu>
Subject: 615 PAPER 65
Message-ID: <Pine.GSO.4.05.10111290338060.6948-100000@kiki.ece.cornell.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


Review of: "Herald: aChieving a Global Event Notification Service," by
L.F. Cabrera, m.B.Jones, M. Theimer

Panagiotis Papadimtratos papadp@ece.cornell.edu

The authors present a hybrid of a position paper and a list of design
directions and goals for a distributed, scalable event notification
scheme. The basic concept that provides for the distribution of the system
is the 'Rendez-vous point' (RVP) and differentiates this design from other
centralized messaging/event notification services. In addition,
simplicity, traded for guarantees on the notification delivery, is a major
goal for the system that targets global deployment.

Herald servers are to reside in numerous machines and publisher and
subscriber client processes create and communicate throught the RVP's.
Clients create RVP's, events are published and the RVP notifies the
subcribed processes for corresponding newly published events. RVP's may be
replicated and reside in different Herald machines in order to increase
fault tolerance. Soft state is maintained in each RVP (i.e., state becomes
obsolete unless explicitly updated) via 'time contracts' and the 'history'
is proposed as a means to deal with periods of disconnection and loss of
state. 

The extreme flexibility in the creation of RVP's and the absence of any
naming scheme raises the issue of resource location; is this the reason
for their proposing the use of agents as the a way to retrieve
notifications (with administrative RVP's also dealing with changes)? The
replication could indeed contribute to load balancing and thus scaling,
but this would also require an explicit way for subscribers to switch from
one replica to another, which is not straightforward at all, especially
because processes that are assumed to act autonomously have to motivated
to do so. Finally, the claim about 'mutually untrusted domains' recurs but
apart from a generic statement about Access control policies to be
employed (by herald servers? at RVP's?) there is no further explanation on
how this parameter is taken into account in the design (or is it merely a
re-phrasing of 'open, distributed system'?)


From mh97@cornell.edu Thu Nov 29 10:52:32 2001
Received: from postoffice.mail.cornell.edu (postoffice.mail.cornell.edu [132.236.56.7])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATFqWT01290
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 10:52:32 -0500 (EST)
Received: from mars (syr-66-24-28-66.twcny.rr.com [66.24.28.66])
	by postoffice.mail.cornell.edu (8.9.3/8.9.3) with ESMTP id KAA20669
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 10:52:31 -0500 (EST)
From: "hao ming" <mh97@cornell.edu>
To: <egs@CS.Cornell.EDU>
Subject: 615 PAPER 65
Date: Thu, 29 Nov 2001 10:52:20 -0500
Message-ID: <000701c178ed$d4703430$6901a8c0@mars>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="US-ASCII"
Content-Transfer-Encoding: 7bit
X-Priority: 3 (Normal)
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook, Build 10.0.2627
X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000
Importance: Normal

Herald: Achieving a Global Event Notification Service
BY Luis Felipe Cabrera

this paper presents the design ideas of a global event notification 
system, hERALD,  which can be the base of many currently popular 
distributed services such as instant messages. on contrast to the 
current centralized system, Herald consists of a bunch of federated 
servers cooperating with each other in a decentralized way which, the 
paper claims, can give it the scalability like the current internet.

In my opinion,  Herald is not a pure peer-to-peer system because of
those federated servers. it is more like a multiple servers + clients
architecture. 

the basic architecture of Herald is as follows:
1. system consists of three building blocks: publishers which generate
   events; clients which subscribe specific events; a Rendevzous point 
   which is set up by clients and maintains subscription relationship 
   between publishers and clients.
   events.
2. event publishers send events to corresponding rendezvous points which
   further send events to subscribing clients.


Herald's main design goals are 
1. scalability. 2. heterogeneous federation. 3. resilience 4 security
5. timliness 6. partitioned operation. 7. support of disconnection 8.
self-administration.  these ithink are basic requirement for a system
like Herald. one interesting aspect is that Herald pushes many 
functionalities to hige layers such as naming service, filtering and 
in-order-delivery. then i am no sure the properties or goals of Herald 
can still be achieved when those services are added on top of it. 

in order to increase fault torlerance, assumption that nodes are not 
reliable is adopted. some compromises are also made to increase the
scalability of systems like states are mainted in a soft-consistent 
way.

the main design techniques are:
1. replication for scalability and fault torlerance.
2. overlay networks for multicasting.
3. time contracts are associated with replica states.
4. event history can be bound by time contract and server rejection.
5. administrative rendezvous points for uniform control messages
exchange.
 
the main contribution of this paper is combining existing techniques to
create a new, globally scalable system. but this paper just discusses
some
design ptinciples. no results are presented. guess it is still in
development.

final comments: i doubt decentralized approach. as it is proved,
centralized 
servers can support the traffic like that going through Yahoo.com. in
fact,
those servers usually consists of many computers, clusters. i do think
they
will have scalability problems. further, centralized approach is much
easier
to develop, much less overhead traffic and much easier maintainance. 

-ming

From daehyun@csl.cornell.edu Thu Nov 29 11:43:49 2001
Received: from wilkes.csl.cornell.edu (wilkes.csl.cornell.edu [132.236.71.69])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATGhnT10541
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 11:43:49 -0500 (EST)
Received: (from daehyun@localhost)
	by wilkes.csl.cornell.edu (8.9.3/8.9.2) id LAA70227
	for egs@cs.cornell.edu; Thu, 29 Nov 2001 11:43:44 -0500 (EST)
	(envelope-from daehyun)
From: Daehyun Kim <daehyun@csl.cornell.edu>
Message-Id: <200111291643.LAA70227@wilkes.csl.cornell.edu>
Subject: 615 PAPER 65
To: egs@CS.Cornell.EDU
Date: Thu, 29 Nov 2001 11:43:44 -0500 (EST)
X-Mailer: ELM [version 2.4ME+ PL54 (25)]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

This paper presents Herald which is a highly scalable global event
notification system.

The main issue in designing Herald is scalability.
Event notification systems are already used in many applications
such as instant messenger and stock price tracking systems. But,
conventional event notification systems are designed to be used in
localized setting such as a building and a campus. With advent of
generalized e-Commerce, global event notification systems are
required which can be used in the internet-scale distributed
applications. Herald is intended to transparently scale in all
respects - the number of subscribers, publishers and events - and
interconnect dynamically changing sets of clients and services.

Herald has a simple model. Event is a set of data items provided at
a particular point in time by a publisher for a set of subscribers.
Each subscriver receives a private copy of the data item by means
of a notification message. Herald does not interpret the contents
of the event data. The operations sequence is as follows;
   1) A Rendezvous Point is created by a client. 
   2) Clients subscribe the rendezvous point. 
   3) A client publishes an event to the rendezvous point.
   4) Herald sends the subscribers the event.

Herald adopts some ideas from the Internet and the Web. Herald
assumes neither that component failure is unusual, nor that the
states are global and always correct. Specifically, Herald makes
the following decisions; First, Herald peers do not assume the
correct behaviors of each other. Second, all distributed state is
maintained in a weakly consistent soft-state manner and is aged.
Third, All distributed state is incomplete and is often inaccurate.
The above decisions help Herald be scalable and robust.

I think the main contribution of this paper is that Herald extends  
conventional notification systems to a internet-scale global systems.
I think its design decisions to achieve scalability are good.
However, because this paper is very short and abridged, they didn't
present specific descriptions. There might be a follow-up paper with
more detailed explanations. 

From jcb35@cornell.edu Thu Nov 29 11:53:17 2001
Received: from travelers.mail.cornell.edu (travelers.mail.cornell.edu [132.236.56.13])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATGrHT12496
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 11:53:17 -0500 (EST)
Received: from travelers.mail.cornell.edu (travelers.mail.cornell.edu [132.236.56.13])
	by travelers.mail.cornell.edu (8.9.3/8.9.3) with SMTP id LAA04997
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 11:53:14 -0500 (EST)
From: jcb35@cornell.edu
Date: Thu, 29 Nov 2001 11:53:14 -0500 (EST)
X-Sender: jcb35@travelers.mail.cornell.edu
To: egs@CS.Cornell.EDU
Subject: 615 PAPER 65
Message-ID: <Pine.SOL.3.91.1011129113739.1085B-100000@travelers.mail.cornell.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

This paper described herald which details the design of a global event 
notification service.

In the Herald event frame, an event refers to a set of data items 
provided at a particular point in time by a publisher for a set of 
subscribers.  Each subscriber wants to receive a notification message of 
particular events.  Herald uses a Rendezvous point, where all event 
publications are sent and to which clients subscribe.

With respect for scaling, when a rendezvous point starts causing too much 
traffic, herald will move some or all of the work for that point to 
another machine.   I am not exactly clear how this will help in all 
cases, say when a lot of subscribers are listening for one popular event 
which occurs at many publishers - how do you pick and choose who to move, 
when moving any will cause some events to go unheard at some publishers?

To improve herald's fault-tolerance, it will replicate some state (for 
example, who is subscribing to what and who is publishing what) at a few 
machines.

Herald uses either unicast or local reliable multicast communications to 
send event notifications out, depending on which is available and more 
efficient.


Herald proposes an sound method for providing global event notification, 
however I would think the naming mechanism would be key in any global 
event notification system.  

From ranveer@CS.Cornell.EDU Thu Nov 29 11:58:33 2001
Received: from exchange.cs.cornell.edu (exchange.cs.cornell.edu [128.84.97.8])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATGwXT13487
	for <egs@popsrv.cs.cornell.edu>; Thu, 29 Nov 2001 11:58:33 -0500 (EST)
Subject: 615 PAPER 65
MIME-Version: 1.0
Content-Type: text/plain;
	charset="utf-8"
Date: Thu, 29 Nov 2001 11:58:33 -0500
content-class: urn:content-classes:message
X-MimeOLE: Produced By Microsoft Exchange V6.0.4712.0
Message-ID: <706871B20764CD449DB0E8E3D81C4D430232E6BD@opus.cs.cornell.edu>
X-MS-Has-Attach: 
X-MS-TNEF-Correlator: 
Thread-Topic: 615 PAPER 65
Thread-Index: AcF49xQuOK3wEHjCR7ium3e5GNvziw==
From: "Ranveer Chandra" <ranveer@CS.Cornell.EDU>
To: "Emin Gun Sirer" <egs@CS.Cornell.EDU>
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from base64 to 8bit by sundial.cs.cornell.edu id fATGwXT13487

Herald: Achieving a Global Event Notification Service
 
Cabrera, Jones, Theimer
 
This paper looks into a new research topic: building a global event
notification service.   Commercial products for these applications use a
centralized approach, with server farms implementing the central server
to which all published events are sent.  The centralized server then
routes these event details to the subscribers.  The main problem with
these schemes is the scalability.  Other problems include the absence of
disconnected operation and the huge message overhead.   Herald tries to
build a global event notification service by targeting several goals:
permitting heterogeneous clients, scalability, resilience, security,
timeliness, and self-administration.  Another key aspect of Herald is to
relegate the implementation of stronger properties such as in network
filtering and consistency to the higher layers.  Therefore Herald aims
to provide the Base layer on top of which several stronger protocols can
be implemented. 
 
Herald uses a Rendezvous Point, which is similar to some of the
multicast protocols that have been proposed for the internet(DVMRP).
The Rendezvous Point is replicated to ensure scalability.  Disconnected
and partitioned operation is ensured by permitting a weak consistency
among the replicas.  A reliable multicast protocol is used to send the
event notification to the subscribers.   This federated approach as
opposed to a centralized one is the research focus of Herald.  There are
a number of research issues that have been outlined and should provide a
good guideline for building the targetted event notification system.
 
The design of a scalable reliable multicast protocol is an area of
current research.  We know that SRM and RMTP do not scale beyong a few
thousand of nodes.  Probabilistic protocols give a better performance;
however, it would depend on the scalability of the underlying multicast
protocols, such as the MBone.  The scalability of MBone to millions of
users is unknown and would be interesting.

From avneesh@csl.cornell.edu Thu Nov 29 12:39:09 2001
Received: from capricorn.ds.csl.cornell.edu (capricorn.csl.cornell.edu [132.236.71.92])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATHd9T20484
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 12:39:09 -0500 (EST)
content-class: urn:content-classes:message
Subject: 615 Paper 65
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Date: Thu, 29 Nov 2001 12:42:12 -0500
X-MimeOLE: Produced By Microsoft Exchange V6.0.4712.0
Message-ID: <97C142C1212ED545B0023A177F5349C40A0A15@capricorn.ds.csl.cornell.edu>
X-MS-Has-Attach: 
X-MS-TNEF-Correlator: 
Thread-Topic: 615 Paper 65
Thread-Index: AcF4/S192b8QDVHfTeODk0GY1JDqjw==
From: "Avneesh Bhatnagar" <avneesh@csl.cornell.edu>
To: <egs@CS.Cornell.EDU>
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from quoted-printable to 8bit by sundial.cs.cornell.edu id fATHd9T20484

Herald: Achieving a global event notification service

Summary/Critique:

This paper presents a design overview of a global event notification
system, which is aimed to be highly scalable spanning dynamically
changin sets of clients and services. The main aim is to tolerate the
uncertainity and lack of trust as exists in the internet. This is
achieved through the design of a rendevous point, which co-ordinates
event subscription and publishing for multiple clients. This rendevous
point is replicated across several nodes to provide greater fault
tolerance.

In terms of performance goals, the paper attempts to provide a service
targeting heterogeneous clients, robustness and security as well as self
organization. The Herald design also chooses not to make naming a
research criteria, this would be interesting to see, since naming
becomes a considerable issue in the design of a system such as this.
There are also some other salient points that might have to be
considered:
1. Event notification: Although herald establishes the property of soft
state maintanence and history mantainence, details of the event
notification mechanism is lacking. There is the issue of asynchronous vs
synchronous event notification to the clients. Under the synchronous
case, the client has to be alive to receive corresponding events, but in
the asynch case (which is required for scalalbility in any case), the
rendevous point may need a pending event queue for clients which have to
be notified. This raises subtle issues such as change in view of the
system, which would require mantainence of this queue. Under a global
notification service, this might not scale, which means that there are
restictions on event notification guarantees.
2. Agent co-ordination for complex queries
3.Finally there was'nt much of a design description as to how and what
security policies( if any) would be enforced.

From teifel@csl.cornell.edu Thu Nov 29 12:45:26 2001
Received: from disney.csl.cornell.edu (disney.csl.cornell.edu [132.236.71.87])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATHjPT21594
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 12:45:25 -0500 (EST)
Received: from localhost (teifel@localhost)
	by disney.csl.cornell.edu (8.11.3/8.9.2) with ESMTP id fATHjKv49424
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 12:45:20 -0500 (EST)
	(envelope-from teifel@disney.csl.cornell.edu)
X-Authentication-Warning: disney.csl.cornell.edu: teifel owned process doing -bs
Date: Thu, 29 Nov 2001 12:45:20 -0500 (EST)
From: "John R. Teifel" <teifel@csl.cornell.edu>
To: <egs@CS.Cornell.EDU>
Subject: 615 PAPER 65
Message-ID: <20011129110610.A38847-100000@disney.csl.cornell.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


Herald:

Herald is a highly scalable global event notification system at Microsoft
Research.  It is a scalable, distributed system that can handle arbitrary
number of subscribers, publishers, event subscription points, and event
delivery rates.  It is fault-tolerant, avoids maintaining most globally
consistent states and will scale to millions of users.

Herald attempts to provide a distributed implementation for global event
notification, which must be real-time.  Current notification systems such
as IM, chat, stock quotes, multi-player games use centralized systems to
support the publish-subscribe information model.  The authors do not
believe that any centralized scheme can be a truly global solution, and
that only a federated approach with multiple, mutually suspicious parties
in different domains of trust that interoperate with each other is
feasible on a global network.  The interesting thing about this is that if
Herald turns out to be a feasible global event notification service, it
will diminish the importance of Microsoft's own MSN IM, etc. network
services.

This paper appears simply to be a design overview paper and presents no
results, performance measurements, simulations, etc.  No hard basis for
evaluating whether this is a good or bad idea and whether it is more
advantageous than the current centralized systems--which are working just
fine at this current time.

From andre@CS.Cornell.EDU Thu Nov 29 12:53:02 2001
Received: from sundown.cs.cornell.edu (sundown.cs.cornell.edu [128.84.96.20])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATHr2T23353;
	Thu, 29 Nov 2001 12:53:02 -0500 (EST)
Received: from khaffy (mail@dhcp99-178.cs.cornell.edu [128.84.99.178])
	by sundown.cs.cornell.edu (8.11.3/8.11.3/R-3.6) with ESMTP id fATHr1206728;
	Thu, 29 Nov 2001 12:53:01 -0500 (EST)
Received: from andre by khaffy with local (Exim 3.32 #1 (Debian))
	id 169Vm5-0005bJ-00; Thu, 29 Nov 2001 12:19:05 -0600
Date: Thu, 29 Nov 2001 12:19:05 -0600
From: =?iso-8859-1?Q?Andr=E9?= Allavena <andre@CS.Cornell.EDU>
To: egs@CS.Cornell.EDU
Cc: andre@CS.Cornell.EDU
Subject: 615 PAPER 65
Message-ID: <20011129121905.A21523@khaffy>
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
User-Agent: Mutt/1.3.23i
Sender: =?iso-8859-1?Q?Andr=E9_Allavena?= <andre@CS.Cornell.EDU>

Herald:

A good introduction to the design of a p2p system, pinpointing down the
interesting issues (scalability, none mututal trust; resilience,
self-organizing, etc.)

Besides that, there is not much interesting going on since they do not
enter the details of any of the issues involved, especially how
would/can work a weak consistent distributed state.

They also say that "If it is broken, don't bother fiwing it" [about the
web] it is not clear but they might say that theyu wont to have to
overcomming of failure handled on the client side.

-- 
Andr� 

From samar@ece.cornell.edu Thu Nov 29 13:21:47 2001
Received: from memphis.ece.cornell.edu (memphis.ece.cornell.edu [128.84.81.8])
	by sundial.cs.cornell.edu (8.11.3/8.11.3/M-3.7) with ESMTP id fATILlT28862
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 13:21:47 -0500 (EST)
Received: from photon.ece.cornell.edu (photon.ece.cornell.edu [128.84.81.138])
	by memphis.ece.cornell.edu (8.11.6/8.11.2) with ESMTP id fATIIqM22491
	for <egs@cs.cornell.edu>; Thu, 29 Nov 2001 13:18:52 -0500
Date: Thu, 29 Nov 2001 13:18:55 -0500 (EST)
From: Prince Samar <samar@ece.cornell.edu>
To: <egs@CS.Cornell.EDU>
Subject: 615 PAPER 65
Message-ID: <Pine.LNX.4.33.0111291317190.31705-100000@photon.ece.cornell.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


Herald: Achieving a Global Event Notification Service

This paper presents the requirements and design overview of Herald, a new
global event notification system being developed at Mircosoft Research. At
each Herald server, Rendezvous Points are created to which subscribers and
publishers associate themselves to receive notifications and publish
events respectively.

The design criteria for Herald are as follows. It is assumed that Herald
will be constructed based on interaction between machines within
cooperating but mutually suspicious domains of trust. The system should be
able to scale without bounds on the internet and should be fault tolerant
towards disconnected, malicious or corrupted participants and should
provide access control. The system should be able to independently
reconfigure and adapt to changes without the need of any external
intervention. The events need to be communicated in a timely manner and
they should be cached for temporarily disconnected clients. Network
partitioning should not affect the operation within each partition.

Herald draws inspiration from the development of the internet, applying
the lessons learnt to its design. Herald peers treat each other with
mutual suspicion and do not depend on the correct behavior of any single
peer. All distributed state is maintained in a weakly consistent
soft-state manner and expired after a time-out interval. All distributed
state is incomplete and may be inaccurate.

For scalability and load distribution, the rendezvous points may be
replicated. This helps in fault tolerance too in the event of some server
going down or getting disconnected. In order to reduce redundant
transmission, Herald implements event notification by means of
multicast-style overlay distribution networks. All the knowledge about
clients, rendezvous points and their replicas are maintained in a soft
state, timer based manner. Event history is maintained in a similar
manner. Also, administrative rendezvous points are present in the network
which notifies interested parties about changes to a rendezvous point. The
authors mention a number of still unresolved issues which need to be
studied in greater detail for Herald to be successful.