drclue wrote:
I thing the most annoying one relates to the fact that many
browsers treat an applet as a plug-in of sorts
Taking into account that NN4 is gone way ago and Microsoft JVM joined
it too, that is the only situation you can expect on a wide run with
Java used on a web-page: just another plugin injected over <objecttag
(leaving out server-side Java like .jsp)
I'd like to avoid a la comp.lang.java.advocacy discussion here, but the
question was posed from the point of view of the most basic regular
user "like my grandma" ;-)
As far as latency in javascript chat , this typically has more to
do with the phrasing of the back-end than anything else.
my context servers can even on a $150 computer pump out over
30 fat transactions a second. Far more if all they are doing is those
tiny chat lines.
The killer with the "Comet" hack is not the per-connection load - it is
negligible. The problem is with the number of simultaneously open
connections. With a more-or-less intensive use and especially with a
sub-optimal server-side processor you'll easily get an analog of DOS
attack on your server - initiated by yourself.
As a reminder: "Comet" hack is based on the age old client spoofing
trick once used for server-side driven image animations on the page.
The core principle is to make the client think that it's downloading
some extremely large file so it would keep the channel open; you also
have to keep some minimum activity over the channel so do not let UA to
"loose the patience" with the "Connection timed out" error. That is
really all wisdom of "Comet". At the pre-historic era one used false
(enormously huge) file size reported for .gif file with server-data
overriding the same scanlines on the picture. With "comet" it is either
an artificially created "border line slow" downstream for say frame or
iframe; or the rather recent miltipart/override MIME in the ajaxoid
response.
Personally I consider all of this *now* as a manually implanted server
virus, but I may be missing the progress curb in this aspect.
In application to web-chats I see nothing wrong with an ol'good ajaxoid
making HEAD request every say second and making GET/POST request as
soon as server reports new data available. That is times more resource
saving, and with all workarounds involved for "Comet" the delay will be
nearly shorter.
Again: maybe I'm just missing the progress here.