Re: htdig 4

ing. J.E.J. op den Brouw (
Thu, 28 May 1998 23:45:09 +0200 (MET DST)

On Thu, 28 May 1998, Andrew Scherpbier wrote:

> ing. J.E.J. op den Brouw wrote:
> >
> > On Thu, 28 May 1998, Andrew Scherpbier wrote:
> >
> > > I might as well share my plans with the list:
> > >
> > > You may or may not like it...
> > > ht://Dig 4 will be written in Java.
> >
> > Ouch!! I hope they'll give me a Pentium <VeryMuchMHz> MMX/II and a lot
> > of memory. The idea is o.k. though......
> Are you talking from experience or do you believe the media hype?

Well, I tried to run the ICQ software for Linux written in Java and
I needed some extra RAM to stay in business. When I compile just
over 10 lines of java code (I'm trying to get the hang of it now),
it takes my machine (P75/16MB) about 20 seconds to compile. C is

When you look at java applications beeing started (I don't mean
applets), most java VM's take up a lot of memory. It is also
an interpreter system, which mean slow. Native code is faster,
and probably take up less memory. But java is more portable...
What will be the performance on htsearch-ing when 10 requests
have to be handled at the same time...
Htdig-ing isn't the problem. It may take up more time to finish, but
the searcher must be fast. Even when only one search request is to
be handled, a response must come within a few seconds (< 20 seconds,
if I believe our students)

> I don't want to go into something that should be discussed in
>, but...
> All the talk about "java is slow", "applets take to long to download", etc. is
> not relevant to this application. Java is actually very well suited for this
> type of task.
> As you may know, I started Contigo Software as a Java only software company.
> We have been selling our server software for over a year and we have yet to
> run into performance problems for which Java is to blame.
> We have about 30 servers running on our main production machine. Granted, the
> machine is a dual 200MHz ultrasparc, but I'm talking 30 servers... The load
> average on the machine normally stays below 2. Oh, did I mention that we are
> *not* using a JIT for any of these servers?!
> Performance problems we *have* run into are mostly due to limitations in the
> OS or hardware. (NT on a dual 400MHz PII became unusable with more than 750
> *active* threads. Solaris did just fine on a single 200MHz ultrasparc...
> Solaris has this nice hidden limit to the total number of TCP sockets that the
> machine can have open... etc.)
> --
> Andrew Scherpbier <>
> Contigo Software <>

This archive was generated by hypermail 2.0b3 on Sat Jan 02 1999 - 16:26:19 PST