Re: ruby as slow scripting language vs java?

[boud]

hi dmitry :)

On Tue, 30 Aug 2005, Dmitry Borodaenko wrote:

> On Mon, Aug 29, 2005 at 10:56:35PM +0200, boud wrote:
> > ruby is a scripting language, not a compiled language, and hence runs
> > much slower than a language like java.
>
> This is a subject of much heated debate in many communities, including
> Ruby community, scripting language community, and computer language
> community as a whole.

Well, with only two of us active here, hopefully this won't turn into
a flamer war :P.

> Formally, this statement is false on several accounts, most obvious

Good. :)

> being that 1) you can only compare applications and not languages, 2)
> not all Ruby applications are slower that similar Java applications, 3)
> speed difference has very little to do with scripting vs. compiled
> languages.
>
> Not to mention the argument that no so long ago was used in defence of
> Java vs. C++: when application is fast _enough_, it's development time
> that matters, not run time.

i definitely agree with this: as long as samizdat is fast _enough_ for
typical uses (i.e. indymedia), then the development time and
possibility to evolve and scale up is much more important.

Also one of the "Publish/Can:" development goals for samizdat:
http://savannah.nongnu.org/cgi-bin/viewcvs/samizdat/samizdat/doc/design-goals.txt?rev=1.99

is "- full p2p hosting and mirroring (Freenet)" which would lead to
CPU time spread across many servers and hence will naturally speed up.


> > According to this analysis:
> >
> > http://shootout.alioth.debian.org/benchmark.php?test=all&lang=java&lang2=ruby&sort=fullcpu
> >
> > java runs many codes anywhere from 50 to 200 times faster than ruby,
> > though it also uses typically up to 5 times as much memory.
>
> The Shootout is well-known for being completely unscientific and having
> ~0% intersection with real life. It says so on its front page: "few
> dozen flawed benchmarks", "we measure particular programs".

Sure.

> For another, also unscientific, but more real-life comparison, check
> this:
>
> http://www.relevancellc.com/blogs/?p=31

OK.

> The interesting fact is that the author is a Java expert and has written
> a book on the Spring framework, which makes it even more astonishing
> that his first try with Ruby seriously outperforms his Java/Spring
> application.
>
> > Would there be any sense to think of preparing a java version (but
> > designed to run on the free versions of jvm's/jre's, not the sun
> > versions) ?
>
> No.

Good, so i won't volunteer :).

> > Personally, i don't have any experience programming in either of these
> > languages. (i've started in C, in the past i wrote in fortran.)
>
> You should try both. This will make you understand why I say strong "No"
> to Java :)

If i get samizdat running, i'll probably try learning some ruby, and
i'm happy to trust others on the problems with java. :)

> > Could it be that none of these tests are for the sort of operations
> > which are slowest for a CMS, e.g. string operations? e.g. sum-file
> > seems to be a program which operates and strings, and it is only about
> > 10 times slower in ruby than java, according to the table.
>
> The Shootout can't be helped, just forget about it. The Spring vs. Rails
> example shows perfectly that most performance gains do not come from the
> language of implementation (or implementation of the language :), but
> rather from coding practices and specific optimizations. In the case of
> Samizdat, I had 10x and even 100x performance improvements (in
> real-life, not some narrow benchmarks) mostly due to better database
> indexing and smarter caching. Last time I checked, about 30-70% of CPU
> time during stress test was spent in PostgreSQL, not in Ruby
> interpreter.

OK, good.

> > i have to admit that except for servers with tight memory constraints,
> > this is a rather discouraging argument against samizdat :(.  Please
> > correct me if i'm wrong...
>
> If I haven't convinced you yet, here is another argument in favor of
> Samizdat: current snapshot of Samizdat contains ~4800 raw source lines
> of *.rb code (counted with wc -l, including empty lines and comments).
> Same count for DadaIMC version that I have gives ~38000, for SF-Active
> it's ~25000, I can't get a hold of a source code snapshot for Mir, last
> time I've seen it the tarball was couple of times bigger than SF-Active.

The source/ directory in mir-1.1 has over 44000 lines according to:
wc -l source/* source/*/* source/*/*/* source/*/*/*/*

wc -l * source/* source/*/* source/*/*/* source/*/*/*/* gives 45000 lines,
so most lines are in the source/ tree.

> And Samizdat code is clean and well commented, you can't say that it's
> short because it's unreadable. And complexity of code grows as an
> exponent of its size. And Samizdat has more that 1/5th of features of
> these two engines (more like somewhere between 1/3 and 2/3 depending how
> you count).
>
> Does this help to decide which code base you'd prefer to deal with as a
> developer? ;-)

Well, no promises yet ;). i'll go through your response to my feedback 
and let's see whether i can get a full installation set up...

Definitely, 
http://savannah.nongnu.org/cgi-bin/viewcvs/samizdat/samizdat/doc/design-goals.txt?rev=1.99
i extremely enticing.


Ah: i still have a question (which will probably be answered anyway if
i succeed in installing): can people publish without having to become
members? i.e. as some sort of guest? The whole idea is that the "entry
barrier" to users (not just developers :P) should be low.

Anyway, i'll go through your response on my installation attempt...

pozdr
boud