P2P-Based Best Resolution Match

Image Access, Retrieval and Delivery

P Contreras, S Johnstone, F Murtagh

CS, QUB

The Multimedia Data Bottleneck

• No matter what capacity our networks have, they are not scalable for access to, and delivery from, databases of very large images and other multimedia objects.

• Wireless is not limited in capacity, but there are other obstacles there.

• Solution: (i) replication, (ii) distribution of servers, and (iii) distribution of services.

• Our objective: effective P2P solutions for collaborative analysis of scientific image stores.

P2P Solutions to the Bottleneck

• On condition that registry services are limited, peer to peer (P2P) provides a viable scalable and robust platform for access to, and delivery from, multimedia data stores.

• Open source JXTA suite of P2P protocols, www.jxta.org, from Sun Microsystems, is used in this work. (Currently some 11500 JXTA developers are registered.)

Benefits of P2P Access and Delivery

• Data and information coding traditionally applied to single files. (E.g. compression). But access is very important too.

• Doyle and Carlson (resp. Caltech, SD) have studied both resolution-based file coding and access patterns. They characterize the complex system which results as one of “highly optimized tolerance”.

• See Murtagh, Starck & Louys, “Distributed visual information management in astronomy”, in Computing in Science and Engineering, 4, 14-23, Nov/Dec 2002, for further discussion.

Principle of Resolution-Based Matching

• For various objectives (e.g. quicklook, correspondence with already available data of given resolution, limited storage space, limited network access bandwidth), we require image data at less than best resolution.

• We will therefore search for ability of an image server to respond to our query for an image of user-specified resolution.

• Image compression based on the Haar à trous wavelet transform provides rigorous flux conservation in image retrieval at a lower spatial resolution.

P2P Retrieval Algorithm

• Client sends getContentRequest• Client receives result, decompresses and

displays. • getMessageProcessor is used to process a

received message from the JXTA pipe. It is notified of a message by Search listeners.

• Server decompresses the image at the required resolution, then recompresses the result, and sends to the client.

How Does JXTA Work?

• We have: provider peers, consumer peers and rendezvous peers. The latter are used in discovery of appropriate resources.

• Files on a server (i.e. provider) have their content description shared in an XML-based advertisement. Providers therefore register a description of their query space and predicates with a hub.

• Groups of peers can be defined with restricted access.

• JXTA uses TLS (similar to SSL3.0) for secure data transfer.

User at server side has clicked on AddContent and is prompted to enterdescription.

Shows the fileshared with itsadvertisement.

Dialog box for sharing the file.

User at client side has searched for the file and entered the parameters for decompression(mr_decomp) and   conversion (im_convert). (Search performed within secure group AstroImageShare)

Shows the dialogfor creation of a new group. In this case, a secureone.

Client joining group. Must enter the correctpassword.

Problems & Concerns

• Frequent churn of JXTA code. We used CMS (Content Management System).

• Search capability, e.g. JXTASearch, is incompatible with recent JXTA releases. Search functionality is important. So far, we have built our own, using pattern matching in the CMS code.

• We have also used the InstantP2P graphical user interface. Updates to code have been made by us in both CMS and InstantP2P.

• Through the use of router peers, firewalls can be traversed via http. This is unlike e.g. Napster which has problems in this area. (By default JXTA uses port 9700 for tcp and port 9701 for http.)

Balance Sheet

• Proof of concept for our application.

• P2P offers dynamic provision of data and information.

• Distributed, and hence robust and scalable, availability of data and information.

• Peers rather than clients and servers.

Future Plans

• Metadata search functionality, including SQL query support.

• P2P may be used for sharing files, as described here, and also for sharing services. E.g. object finding with return of a results table.

• Use of JXME, JXTA Micro Edition (maybe availing of a relay peer) for deployment on small devices.