- From: Alexandre Rafalovitch <alex@access.com.au>
- Date: Fri, 22 Nov 1996 10:28:37 +1000
- To: JEEVES-INTEREST@JAVASOFT.COM
- Cc: www-jigsaw@w3.org, Anselm Baird-Smith <abaird@w3.org>, michael.oconnell@javaworld.com, chuckr@postoffice.ptd.net, daniel@vpro.nl, soak@sigs.com
If you are on CC list, please scroll down to Competition Offer. I included you because I thought you might be interested. Sorry if you are not. At 3:32 AM +1000 22/11/96, Steve <daddyo@SPARKY.OROAD.COM> wrote: >> Ps. I admit being biased toward Jigsaw and my main reason on this list is >> because I am trying to make Servlet support for Jigsaw (some of you might >> be interested in this, anybody?). However, I am trying to be very objective >> :-} and would be grateful if anybody can correct me where I am wrong. > >Great summary of Jigsaw! One other con I might mention is the fact that >when you >load a new resource you can't 'unload' it. That is, short of either restarting >the server or recompiling your class with a different name (we've been >going with >MyResource1, MyResource2, etc.) and adding it again. I've been told that this >is due to a bug in the JDK ClassLoader. Although this sounds like a real pain, >surprisingly it hasn't really hampered us that much. I haven't done nearly as >much with Jeeves, but does it suffer the same problem? It does. Actually, current implementation of Jeeves does even worse. It does not unload Servlet instances themselves either. That means that once your Servlet is called, it is going to stick around until server is restarted. One of the earlier posts was talking about this in relation to bad initialization problems (when you try to establish DB connection and fail, you want to get rid of Servlet because it is useless without DB access). Jigsaw on another side is using LRU resource management with notification of unloading in background thread (sort of like finalize), so resource can take its time closing its connections, etc. and does not have to be in a hurry. I feel that destroy() method of Servlet even when(if) it will be called by Jeeves would not be given background thread but the main thread, therefore possible timeouts,etc. "Stanford S. Guillory" <Stanford_Guillory@PCORDER.COM> wrote: >I would add that it is currently much too slow for any serious use. Have >you done >any comparisons of performance of Jigsaw and Jeeves? At NCSA we definitely >appreciated the features Jigsaw brought to the table, but found it too slow >to be >our "actual" web server. I am doing a lot with Jeeves now, but have not >actually >tried to get a feel for its performance. On the list :*) I agree that Jigsaw is relatively slow. I only want to point several things out. First, Jigsaw is full HTTP1.1 compliant (Jeeves is not) that means quite a bit more processing is going on. Second, Jeeves is still having problems with sending big (>20k?) files. That might be due to IO model of Jeeves. Once they will fix the problem, who knows the performance penalty they will pay. Third, when you say it is slow, did you do performance checking yourself? Was it alpha1 or alpha3. Anselm improved performance significantly for alpha3. Fourth, Jeeves capabilities as implemented now are limited. It has all that bright future with Servlets running around and doing distributed load processing, but the future is not here yet and I am sure the design of Jeeves will change at least twice before they will be able to do that without a serious strain on the server. I am as bright eyed as a next computer hacker, but I am trying to be realistic too, because I am SICK of lies we are fed in all those happy product releases and conferences!!! I propose a competition. ___________ Offer for competition ____________ Let's choose several good examples demonstrating power of web servers (eg, web counters, SSI, auth protected areas vs. non-auth protected areas, on the fly MCF generation of directory content, changing advertisements a la ad.doubleclick.com, capability based page serving, cookie based operations, etc.). Let's implement it (or use existing implementations) in CGI scripts for NCSA server, Servlets for Jeeves and resources/filters for Jigsaw. Then, we will create identical directory structures for all three web servers and do the shakedown test which of them can perform better with all that flexibility. Then we will be able to talk which server is faster. Some problems here I see: Big files problem for Jeeves, in Jigsaw it is much more efficient to write filter sitting on a resource than do /cgi-bin/php.cgi/documents/index.html type of stuff, on other servers it might be the only way, so directory structures might have to be functionally and not physically identical. The other problem is to have somebody contribute a machine on which it all can be tested. WinNT/95 with JITed VM would be a good platform to run java servers. And I assume NCSA runs on WinNT/95 too? I don't expect any problem with pouncing tools. There are several around, many are free. I am ready to create implementation of all the above mentioned problems (and any others) for Jigsaw under CopyLeft license (or some such), anybody can do that for Jeeves? There are many implementations in Perl for normal servers, I believe. All the sources produced could be good examples of cool Jeeves and/or cool Jigsaw demonstrations and [potentially] can help to find hidden bugs in all servers involved. Maybe even somebody of Jeeves team could join in. ___________ Offer for competition ____________ So here it is. I believe it is easy to write a minimal server that does not do much. However, after a while you want to add to it bits and pieces. That might slow it down to crawl or may just be impossible to use at all. On another side, you might have server that is slower for basic services, but is very good with bells and whistles added on. And any webmaster after a while does want to add bells and whistles to his site. Am I right? (I am not a web master myself :-{ ) Anybody is ready to take it on. Anybody can provide a web space to put competition rules, judges results, specs of implementations, sources of implementations? All participants will have a great opportunity to expose their skills to the world and get their name known in internet circles. Opinions? Alex. Ps, I am CCing it to number of people who might be interested. That includes several other java WebServers makers and commercial organizations, that might be interested in it. Members of Jeeves e-mail list, please remove them off your list for general/hate/love discussion. Anybody who is not on the list and want to stay in touch, contact me and I will forward all info when needed. alex@access.com.au
Received on Thursday, 21 November 1996 18:28:06 UTC