Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 15 May 1997 15:49:54 -0700 (PDT)
From:      Alex Belits <abelits@phobos.illtel.denver.co.us>
To:        Terry Lambert <terry@lambert.org>
Cc:        "Russell L. Carter" <rcarter@consys.com>, pgiffuni@fps.biblos.unal.edu.co, hackers@FreeBSD.ORG
Subject:   Re: Cluster Computing in BSD
Message-ID:  <Pine.LNX.3.95.970515151713.6118D-100000@phobos.illtel.denver.co.us>
In-Reply-To: <199705152117.OAA15760@phaeton.artisoft.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 15 May 1997, Terry Lambert wrote:

> > Hey clusters are a fine approach to distributed problems, but parallelism 
> > across physically distributed systems has not succeeded often enough to
> > merit more than the merest blip in the computing industry.
> 
> How many people would class the World Wide Web itself as a successful
> (though asymmetric) distributed cluster?

  Umm... different class of task, it doesn't "scale" to handle anything
but independent services referencing each other for clients. "Internal"
traffic in that system is limited to search engines, but both distributed
and "single large box" approaches coexist among them.

> How about "web farms", which are generally symmetric, but grossly
> load balanced via DNS rotor... do they count as clusters?

  Those ones are designed to limit transfers between them, and at most
have remote filesystem (so all their processing power is used to handle
slow TCP/IP over the internet and convert requests to fast one across LAN
to single fileserver) or, more sane approach, syncronized local file
storage (so they handle TCP/IP and fast local file I/O), or, run the same
set of programs/modules/cgi/... on all boxes (and have rather unusual for
clusters limitation that everything should be done locally).

  I've made HTTP server that can distribute requests between processes
running on other hosts by round-robin'ing and applying rules
(cookies/sessions/...), but I don't think, that's a real cluster
functionality -- while that system requires fast connections between boxes
and large amount of traffic is internal for it, the topology is limited to
star-like one -- other kinds of topology can be in theory used in such
system, but it's incomparably harder, and I've never seen it done that
way.

> How about the DNS services themselves?

  Closer, but still topology limitations and purpose,/design not anyhow
related to high-performance systems that benefit from connections to
increase processing ability. Still good distributed storage/search system.

--
Alex




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.3.95.970515151713.6118D-100000>