AF:
NF:0
PS:10
SRH:1
SFN:
DSR:
MID:
CFG:
PT:0
S:andy.sharp@lsi.com
RQ:
SSV:mhbs.lsil.com
NSV:
SSH:
R:<Brent.Kingsbury@lsi.com>
MAID:2
X-Sylpheed-Privacy-System:
X-Sylpheed-Sign:0
SCF:#mh/Mailbox/sent
RMID:#imap/LSI/INBOX	0	4B6A08C587958942AA3002690DD4F8C3BA7DA6D3@cosmail02.lsi.com
X-Sylpheed-End-Special-Headers: 1
Date: Thu, 4 Mar 2010 21:13:11 -0800
From: Andrew Sharp <andy.sharp@lsi.com>
To: "Kingsbury, Brent" <Brent.Kingsbury@lsi.com>
Subject: Re: x86 SIMD and the tuxfs world....
Message-ID: <20100304211311.05722d43@ripper.onstor.net>
In-Reply-To: <4B6A08C587958942AA3002690DD4F8C3BA7DA6D3@cosmail02.lsi.com>
References: <4B6A08C587958942AA3002690DD4F8C3BA7DA6AA@cosmail02.lsi.com>
	<20100304154101.25194911@ripper.onstor.net>
	<4B6A08C587958942AA3002690DD4F8C3BA7DA6D3@cosmail02.lsi.com>
Organization: LSI
X-Mailer: Sylpheed-Claws 2.6.0 (GTK+ 2.8.20; x86_64-pc-linux-gnu)
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

I'm glad you got a good laugh out of it.  I was feeling rather impish
at that moment.

Seriously though, it would be a big favor for me if you could ask
around the folks a bit.  Before you quite joined, I gave a rather
informal talk on tuxstor, and the folks in B-town at the time dialed
in, and someone, I don't know who, expressed a lot of interest.  So I'm
wondering if that person, be it David Olien, Chris or someone else, is
more interested in jumping on some of this stuff, then that's what we
should do.  Otherwise it will probably end up in your lap.

The code in question is the same code you're working on now, only it'll
be running under the Linux kernel.  So some of the basic facilities
have to be re-implemented.  In other words, we're porting it to Linux.
There are a number of tasks under that heading.  Let me know if you
want to know more.

Once the "resource allocation" has been fully signed off by all the
dancing elephants, Jobi will probably fill you in on some of the stuff.

As for SIMD instructions and their registers, do you think they aren't
currently handled properly by the kernel?

Cheers,

a

On Thu, 4 Mar 2010 17:21:32 -0700 "Kingsbury, Brent"
<Brent.Kingsbury@lsi.com> wrote:

> Andy:
> 
> That reply cracked me up (at multiple points)! Thanks. I did need
> that. And if you ever need a flame broiled Whopper, maybe I'm the
> person to ask. :-)
> 
> Re: tuxstor, I confess I still don't know (exactly) what that is.
> There seems to be a 'tuxfs' branch which (I assume) feeds into
> tuxstor, but until I saw the Orion slides, all I had for a future
> platform architecture was just the discussions back in 'Dec during
> the interview.  There's probably tons of documentation about tuxstor
> which I've not read, and that's the problem.  Catching up.  I am
> utterly incompetent at the moment because I don't know this source
> code.  I am working at that though, and trying to fix the second bug
> assigned to me while pondering a possible future directory layout for
> the filesystem that could handle gazillions of entries without
> falling over.
> 
> In the meantime, I take the answer to my SIMD instruction use question
> as: yeah, we could make a design that allows us to ("in the fullness
> of time" you understand), make good and clever use of the SIMD
> instructions to speed lookups.
> 
> As for what to ultimately work on, I'm just really grateful to be
> working at a place that seems to actually want to build something.
> Whether that is the filesystem (which I really like), or what you are
> working on (which I would also probably really like), I'm just happy
> to be here.
> 
> Without naming my previous employer who I worked for involuntarily
> because we got bought (and then taken apart), I'll simply say I like
> to work at a place where people are inventing stuff, instead of
> merely having the word: "invent" placed in an empty company
> logo/slogan of that unnamed company I keep talking about.  And now
> that I am remembering that experience again just now, my symptoms
> have returned and I need to increase my medication!
> 
> Thanks, Andy.
> 
> --BK
> 
> -----Original Message-----
> From: Andrew Sharp [mailto:andy.sharp@lsi.com] 
> Sent: Thursday, March 04, 2010 3:41 PM
> To: Kingsbury, Brent
> Subject: Re: x86 SIMD and the tuxfs world....
> 
> On Thu, 4 Mar 2010 16:22:53 -0700 "Kingsbury, Brent"
> <Brent.Kingsbury@lsi.com> wrote:
> 
> > Andy,
> > 
> > 'Sorry to bother, but I've a question I imagine that you (or someone
> > in your immediate circle) could make a judgment upon, which is this:
> > 
> >    Will it be possible for us to use the x86 SIMD instructions in
> > our future tuxfs work when running upon our future Nehalem Xeon
> > x86-based line?
> > 
> > I realize I probably look seriously nuts asking this, but there are
> > some potential optimizations to be had once our C code dust settles,
> > provided that we layout certain data structures in the appropriate
> > ways.   I'm thinking specifically of a possible directory block
> > scheme wherein we lookup pathname component names by hash (which we
> > do already), but wherein all the hash values are in an array (they
> > aren't this way now), and wherein we wouldn't necessarily be wanting
> > to sort the hash values - depending upon the circumstances.
> > 
> > If our 'eee' environment wrapping in the future x86 product
> > environment remains non-preemptive (it appears to me to be that way
> > now from a cursory look), then there would be no need to
> > save/restore the SIMD register context when context switching
> > between threads.  I do however wonder how this would all work in
> > the virtual machine environment of which I saw a recent diagram in
> > the Orion slides.  In that world, I presume the individual virtual
> > machines are preempted ... whenever ... which would seem to
> > necessitate a save/restore of the SIMD register context when
> > switching between virtual machines if more than one virtual machine
> > context had code using those registers....
> > 
> > Thanks for any/all comments and insights,
> > 
> > --BK
> 
> 
> Burger King?
> 
> Anyway, the very short answer is: it would be the kernel's job to
> handle this, so I don't care!  Haha, tra-la-tra-lee.  I should smoke
> less crack, really.
> 
> OK, so maybe you'd like to work on some tuxstor stuff?  I have several
> tasks already with your name on it.  However if you tell Jobama I told
> you, he will come to my office and try to kill me.  If you'd be
> interested in some of that work, let me know, and if there's anybody
> else up there in B-town who would like to work on some tuxstor kernel
> stuff, also let me know.  I'll never tell a soul who spilled the
> beans.
> 
> Normally I would just come around and ask for myself, but they won't
> spring for a plane ticket just now.  And oh yeah, I don't have time
> right now to make a trip to Oregon.  Maybe soon though.
> 
> Cheers,
> 
> a
