Talk:AEOLUS IST/FET Project

From AEOLUS IST Project

Dear Collegues, can you please forward this announcments? Many thanks.

Luigi Liquori

====================================

POST DOC ANNOUNCEMENT at INRIA Sophia Antipolis

deadline 31 march. how to apply: http://www-sop.inria.fr/act_recherche/formation/offres_de_post-doc_sur_sophia_en.shtml#mascotte3

=====================================

Subject title: Grid Computing meets Global Computing and Overlay Computing: Desing, Implementation, Test, and Simulation of a Resource Discovery-Oriented Model using Low Level Network Protocols. Subject description: - Abstract and General Setting. The scope of this Post-doc is to conceive a light-weight infrastructure called, informally, Arigatoni that is suitable to deploy, via the web, the Global Computing Paradigm (recall that a global computing refers to computation over global computers, exploiting their universal scale and the programmability of their services). The aims of this Post-doc is to design, implement, testing and eventually simulate a new light-weight architecture that in principle is suitable to deploy the global computing paradigm using a minimalist SW infrastructure that relies, as much as possible, on the lower levels of the OSI stacks (like e.g. TCP/IP and UDP). Compared to the OGSA infrastructure (e.g. "a la" Globus), the Arigatoni infrastructure is dramatically simpler and exploit the lower level of the OSI stacks. In principle, it could be deployed firstly in an intranet and further from intranet to intranet by overlapping a Global network on the top of THE actual network. In other words: programming a Collaborative Global Web over the plain web. Arigatoni is a lightweight communication model for dynamic Resource Discovery. Inspired by the Publish/Subscribe paradigm, the Arigatoni model implements a Resource-Discovery Oriented Overlay Network. Entities in Arigatoni are organized in Colonies. A Colony is a simple virtual organization composed by exactly one leader, offering some broker-like services, and some set of Individuals. Individuals are SubColonies of Individuals, or basic units called Global Computers. Global Computers communicate by first registering to the Colony and then by mutually asking and offering services. The leader, called Global Broker, has the job to analyze service requests/responses coming from its own Colony or arriving from a surrounding Colony, and to route requests/responses to other Individuals. After this discovery phase, Individuals get in touch with each others without any further intervention from the system, typically in a P2P fashion. Communications over the behavioral units of the model are performed by a simple Global Internet Protocol on top of TCP or UDP protocol. Arigatoni provides fully decentralized, asynchronous and scalable Resource Discovery, that can be used for various purposes from P2P applications to more sophisticated Grid applications. The main focus of this paper is to present the Resource Discovery mechanism used in Arigatoni, along with some simulations that show that Resource Discovery in Arigatoni is efficient and scalable. We briefly detail the recipe of the Arigatoni infrastructure. In the Arigatoni jargon, a Global computer (GC) is a cheap computer device composed by a small RAM-ROM-HD memory capacity, a modest CPU, a >=20 keystrokes keyboard, a >=1.5 inch screen, an Ethernet port, an usb port, and very few programs installed inside (one simple editor, one or two compilers, a sendmail client, a mini browser, a GSM module, etc). At this stage of the design, the OS installed in the GC is uninfluent, while we could envisage, in a later stage, to push later some of the Arigatoni features inside the OS itself. Among the programs in GC the Global Resource socket (GRS) is a small daemon which is devoted to satisfy the request of the GC w.r.t. the global network (typically it could ask for big computational power, i.e. the GRID, or ask for a particular piece of SW, i.e. peer-to-peer, or ask for memory space, or ask to transfer one non completed local run in another GC, saving the partial results, under the case of a catastrophic scenario, like, e.g. fire, terrorist attack, earthquake etc...). More precisely: the request of the global computer is first filtered by a small piece of SW the GRS, embedded in the GC. This piece of SW typically receive a request encoded in a very low-level protocol to minimize overhead. The job of a GRS is to analyze the request and forward the request to the another unit (SW or HW) of the Arigatoni framework: the Global Resource Broker (GRB); the latter is devoted to ack the request of the GRS and find the closest resource available in the intranet (if possible) or delegate the request to another GRB present in another intranet. In principle many GRB can be connected in a hierarchically fashion. The ``algorithmic intelligence of a GRB rely on the well known graph-based algorithmic theory, where the project team Mascotte is leader. Also at this stage, we will try to keep the interaction within those units at a level of the OSI stack as much low as possible. It is the job of one GRB to negotiate the resource demanded by the GC. The resource will be available by one or many other GC's (the servants). In case the GRB will be able to satisfy the request of the (client) GC, it will send directly all the information necessary to make the client able to communicate with the servant; thus, the client will directly forward the request to the GC servants. This task will be done by the GRS of the GC client always using (or programming) a low-level protocol of the OSI stack. Our hope is to minimize the overhead of the classical middleware present in the market. The first approximation of the experiment is to get rid of all security problems, at least in a secure by construction intranet scenario. Security problems will be treated later using common techniques of encryption, like Rsa, security protocols, etc. We envisage at least the following global scenarios: (1) Ask for computational power (i.e. the GRID); (2) Ask for memory space; (3) Ask for bandwidth (i.e. VOIP); (4) Ask for file retrieving (i.e. P2P); (5) Ask for web service (i.e. Google query) - Required Skills. Some of those skills are suitable:Network programming and class loading with Java, TCP/IP, UDP, HTTP, ... protocols, distributed computing algorithm, object-oriented (message passing) paradigm, routing packets, knowledge of NS network simulator tool, performance evaluation techniques (Poisson Law, Queing theory, ...), mechanical proofs tecniques (like knowledge of Coq, Isabelle, ...) - Reading. The interested candidated can have a look on http://www.inria.fr/rrrt/rr-5805.html. This post-doc is in the mainstream of the IST FET AEOLUS - Algorithmic Principles for Building Efficient Overlay Computers. - Tentative Scheduling [month 1-2] brainstorming and sketch of the architecture and inventory of the technology for the light-weight infrastructure Arigatoni. [month 3-6] delivery of the first $beta$-specification of the Arigatoni architecture (included a study of the used protocol to communicate), and start of implementations of GRS and GRB units. [month 6-8] start of testbeds of some scenarios at INRIA Sophia intranet. [month 8-10] this month will be fully dedicated to finish the implementations of the Arigatoni framework and their protocols. [month 10-12] packaging of the SW release, SW documentation, writing of some technical reports and international publications. INRIA Research team: Mascotte Contact: Luigi.Liquori[at]inria.fr, cosnard[at]sophia.inria.fr

--

Luigi Liquori, Ph.D, 
INRIA Researcher, Sophia Antipolis
Project Team Mascotte
Vox : +33 4 92 38 71 93 
Fax : +33 4 92 38 79 71   
Mob : +33 6 65 39 51 32 
Url : www-sop.inria.fr/mirho/Luigi.Liquori
Eml : Let (*,#)=(.,@) in Luigi*Liquori#inria*fr
Pst : INRIA, 2004 Route des Lucioles - BP 93
      FR-06902 Sophia Antipolis, France