To: Frank A. Coluccio who wrote (1895 ) 8/1/1999 9:54:00 PM From: Scott C. Lemon Read Replies (1) | Respond to of 5853
Hello Frank, I wanted to also thank you for you two posts ... very well written, as usual. They caused me to wake from my thinking period to once again interject some comments that I believe are relevant to your latest posts ... As you know, for the last several years I have been closely following the developments in the "proxy/cache" software and hardware areas. Most of this is because I truly believe that most people are missing the significance of the technology that this represents. I believe that this technology is the first step towards, what I call, "Object Routing" and your latest posts have some linkage to these concepts. In your posts you talk about the two models, and how they are represented ... and I agree. But I believe that other people (not you) go too far in "fighting" the rising intelligence of the network ... the infrastructure. Please do not misunderstand ... I'm not a "intelligent network / dumb end-node" proponent ... but I do think that the components of the network will gain some IQ points in the future. For example, the leap from hubs and routers to switches was a very good one, and the "intelligence" of the devices improved. A hub is *much* "dumber" than a switch ... but switches are more desirable due to the improved traffic handling capabilities. Likewise, vendors are now creating equipment that tried to more intelligently "peek" at packets to look deeper at the protocol headers to "switch" or "route" more efficiently ... again more "intelligence". I believe that there are a couple of areas that we might see gaining ground in the network to raise this level of intelligence even more ... and they relate to "Object Routing" or "application protocol routing" or "proxy/cache" technologies. 1. Better alignment between "packets" and the objects that they represent. What I mean by this is that we (as an industry) seem fixated on packets ... but a packet is usually a mere fragment of the object that the end-user is attempting to communicate across the network. Because of this, there are a large number problems introduced, and solutions that we can not provide. Where possible (and there are a large number of places this is possible i.e. HTML documents, small GIF or JPG files, mail messages, etc.) if packets actually became the whole object, then new methods of routing and caching could be employed. On top of this, whole new services could be developed to utilize the object within the infrastructure of the network because it would be a complete object with context. 2. As certain protocols become IETF ratified and are in large use (i.e. SMTP, HTTP, FTP, etc.) they can be transparently redirected without any knowledge of the end-user or their application. This redirection can then be used to route that protocol over different links, or through other hardware or software which could improve performance or routing. Examples of this are now becoming more commonplace, for example, with Cisco's WCCP caching architecture or Novell's ICS/BorderManager caching. Anyone using @Home or AOL has probably experienced transparent redirection of web browsing traffic, and on a trip to Russia I experienced the occasional "Squid Error" while cruising the web ... signs of transparent redirection. 3. Due to the design of the ISO 7 layer model, it becomes possible to replace any of the lower layers with new protocols, or layers, without effecting the upper layer applications. So when my web browser using HTTP is transparently redirected to a cache, I no longer can be certain how that cache communicates to another cache to get the web content. The cache could be using a "whole object streaming protocol" to communicate through a mesh of caches, eventually reaching another edge cache near my desired web site where HTTP would again be used to fetch the page from the server. So instead of "dumb" routers slinging packets from my browser, through the network, to the server and back ... I would have a higher-layer "object" request being forwarded through more intelligent infrastructure (capable of caching and other processing) to a point nearest to the server, and a whole object being efficiently streamed (and possibly cached) back through that infrastructure to me. And the layered architecture allows this to occur without end-user knowledge or changes. I agree that it is not something that will or can be done for *all* protocols and objects today ... but as we continue to learn which ones are "standard" and in wide use I believe that large amounts of traffic can be better optimized than simply throwing higher-speed packet routers at the problem. (And on top of that, an infrastructure that can then provide services which process whole objects is a whole new value add ...) I'm looking for feedback from you on this due to your knowledge and perspective. There are many people who, IMHO, take the "dumb network" issues a little to far ... at which point we would be using repeaters and hubs ... no switches and routers. I am trying, in my research, to take another leap which would be to move beyond the packet orientation to object orientation in networking. To me, packet routers have just been around too long ... there has to be a better way to transport and route common objects ... ;-) Scott C. Lemon