Date: Thu, 12 Dec 2002 10:50:48 -0700 (MST) From: "Ronald G. Minnich" <rminnich@lanl.gov> To: Andy Sporner <sporner@nentec.de> Cc: freebsd-cluster@FreeBSD.ORG Subject: Re: sharing files within a cluster Message-ID: <Pine.LNX.4.44.0212121030540.18020-100000@carotid.ccs.lanl.gov> In-Reply-To: <3DF8B537.70908@nentec.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 12 Dec 2002, Andy Sporner wrote:
> Well let's review here. RPC's have many dependencies for correct
> operation (portmap, etc) regular socket I/O does not--so which one is
> more reliable???
Actually, that's SunRPC, one variation of RPC, that requires all that
stuff. RPC is a very simple concept, and what you typically see is people
implementing things that are actually RPC but calling it socket I/O.
Pretty much any protocol running over sockets that does a request with a
"procedure number" and expects a response to that request is RPC.
You can even do async RPC, sources on my web page to a modified SunRPC
that does this. Performance is pretty good to lots of nodes ... works on
FreeBSD, or used to.
My apologies, the "linuxbios naming" thing really teed me off. I'm also
sorry to be so darned tactless, a lot of those messages I was just
pounding out between Pink bringing firefighting or early in the morning
when I should sleep instead of write email. I apologize to this list.
> Your "read and learn" comment is just the kind of pompousity I expect
> from the academic world.
I gotta watch that. I'm being terse and I guess I should not be. Esp.
since I have not been in the academic world for 14 years, I've been
building HPC systems for various gov't and commercial entities instead.
I'm getting really frustrated with this list, however, since people seem
to be refusing to learn from linux because it's "evil", and at the same
time proposing stuff that's been tried years ago on linux and found not to
work. I think it is really important that the FreeBSD community break out
into something new and really neat. At a minimum, however, I'd like to see
FreeBSD have equal capability to Linux clustering for HPC. And I think
before we condemn technologies ("RPC") it's best to make sure it's the
technology, and not a specific implementation of it, that are the issue.
I would really encourage people to get at the literature of SSI, going
back to Farber's original DCS paper in '72, and try to find an angle that
is somehow new and uniquely suited to FreeBSD. And "Linux sucks, FreeBSD
rules" is not going to be the thing that does it. What about FreeBSD can
somehow make it much better for clustering? That's the problem I've never
been able to answer -- maybe one of you can.
This list is sporadically active, but there have not been tons of new
ideas crossing it -- seems like we see the same stuff over and over again.
> If this was such well plowed ground, why does it not already exists in
> FreeBSD?
Because, and as a former FreeBSD cluster builder I hate to say this, Linux
won. And the mindshare is in Linux, as are the compilers, 3rd party apps,
and all the big vendors. There are many 10s of millions of dollars of
clustering money being spent by IBM, HP, Intel, etc., etc. and absent
something really new and innovative I don't see FreeBSD breaking in.
The closed-minded nature of the FreeBSD core has not helped. I wrote a DSM
for FreeBSD 2.0.5 ca. 1994, which was based on simple mods to NFS --
called MNFS (really it was a port from a SunOS version I also wrote). It
required a simple extension to the VM system (it required the VM layer to
tell the VFS layer if a page fault was for read or read/write -- that's
what SunOS does) and I could not get the FreeBSD core to add this simple
additional parameter -- it violated "information hiding between layers",
which is good in theory but in practice the more info the better. (this
code is still on my web page along with the MNFS papers).
8 years ago both Sandia/Livermore and I were building freebsd clusters.
It's arguably the better OS. Is it so much better or different from Linux
that it is inherently better for clustering? Probably not. So we got
clobbered by the rest of the world. What we got steamrolled by was the
availability of so many 3rd part apps.
> As for comments to you, I think it is pretty clear that when I made an offer
> to port the very thing you asked for a reply is in order.
I'm glad to hear it, sorry for lack of reply. Please get the clustermatic
ISO at www.clustermatic.org and then we can talk about porting. I would
LOVE to see this happen.
You can find an initial private name space VFS implementation for FreeBSD,
called v9fs, I think on my web page.
Sorry, Andy. I'll try to behave.
ron
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-cluster" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.4.44.0212121030540.18020-100000>