Main Page | Report Page

 

  Computers Forum Index » Computer Architecture » Fenceless atomic RMW...

Author Message
Terje Mathisen...
Posted: Tue Oct 12, 2010 11:59 pm
 
Bakul Shah wrote:
Quote:
On 10/11/10 7:03 PM, Rob Warnock wrote:

Microsecond? *MICROSECOND?!?* Nick, remember, a microsecond is a *MILE*
when it comes to GPS. No, the GPS constellation is synchronized to
within a few *nano*seconds.

Rob, how do you map a microsecond to a mile? And a mile where?
I am being lazy (& dense) but I couldn't figure out how. Even
light travels 300 meters in a microsecond.

I think Rob was talking about 5 us, which is close enough to a mile, at
least for US engineers (and track & field runners, hereabouts we prefer
1500m).

Anyway, the speed of light is a pretty good proof of the superiority of
the metric system (300,000 km/s) or even better, the metric A paper
formats: A regular A4 sheet of paper is extremely close to a ns in
length. :-)

Terje

--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"
 
...
Posted: Wed Oct 13, 2010 7:54 am
 
In article <n9vfo7-n0j.ln1 at (no spam) ntp.tmsw.no>,
Terje Mathisen <"terje.mathisen at tmsw.no"> wrote:
Quote:
Andy "Krazy" Glew wrote:
On 10/10/2010 11:16 PM, Terje Mathisen wrote:
I consider cache to be a much wider term than what's usually considered,
in that I believe algorithms is a way to cache insight about how to
solve a particular problem:

Finding a new O(n) or O(n*log(n)) algorithm for something that naively
requires O(n*n), and then implementing it in a library, is the most
effective form of caching there is.

Wow! I've been reading your .sig for years, and never realized you
thought this way.

Haven't I told you this in one of our many late-night discussions?

In fact, my .sig dates from the moment when I suddenly realized that all
programming beyond naive/brute force is a form of caching.

I think I know what you mean, but feel that you are distorting the
term unreasonably. A hell of a lot of things are reasonably viewed
as caching (B-trees, for example), but there are plenty of things
where the solution is to use an equivalent but different mathematical
formulation.

And it does NOT apply to software engineering, which is definitely
part of programming, where the key is to constrain yourself as much
as possible consistent with achieving your objective.


Regards,
Nick Maclaren.
 
Terje Mathisen...
Posted: Wed Oct 13, 2010 10:13 am
 
Andy "Krazy" Glew wrote:
Quote:
On 10/10/2010 11:16 PM, Terje Mathisen wrote:
I consider cache to be a much wider term than what's usually considered,
in that I believe algorithms is a way to cache insight about how to
solve a particular problem:

Finding a new O(n) or O(n*log(n)) algorithm for something that naively
requires O(n*n), and then implementing it in a library, is the most
effective form of caching there is.

Wow! I've been reading your .sig for years, and never realized you
thought this way.

Haven't I told you this in one of our many late-night discussions?

In fact, my .sig dates from the moment when I suddenly realized that all
programming beyond naive/brute force is a form of caching.
Quote:

In fact, I have often had mental debates with my model of you, about
whether it is cheaper to store in a cache vs. recompute.

That's always a valid consideration. When you realize at one point in
time that in the current environment recalculation is the right
solution, then you're really caching that insight.

Just like data caches, such insights can become stale.

Terje

--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"
 
Terje Mathisen...
Posted: Wed Oct 13, 2010 10:25 am
 
nmm1 at (no spam) cam.ac.uk wrote:
Quote:
In article<nareo7-sdh.ln1 at (no spam) ntp.tmsw.no>,
Terje Mathisen<"terje.mathisen at tmsw.no"> wrote:

Anyway, the speed of light is a pretty good proof of the superiority of
the metric system (300,000 km/s) or even better, the metric A paper
formats: A regular A4 sheet of paper is extremely close to a ns in
length. :-)

Nah. A nanosecond is a foot Smile

1 foot: 30.48 cm
A4 paper: 29.7302 cm (From the definition, usually given as 29.7)

1 ns: 29.9792 cm

I.e. 2.5 mm vs 5.0 mm error.

I rest my case. :-)

Terje
--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"
 
Terje Mathisen...
Posted: Wed Oct 13, 2010 11:02 am
 
Rick Jones wrote:
Quote:
Terje Mathisen<"terje.mathisen at tmsw.no"> wrote:
A regular A4 sheet of paper is extremely close to a ns in
length. :-)

I never got to see it, but somehow I find it difficult to imagine
Grace Hopper handing-out sheets of A4 paper... Smile

30 cm lengths of single-strand phone wire is a lot easier to carry
around in some quantity.

Besides, US Letter (as well as your other sheet sizes) aren't even close
to a ns in length.

Had she been European instead otoh, she would never have made it as far
as she did in any national Navy. :-(

Terje
--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"
 
Terje Mathisen...
Posted: Wed Oct 13, 2010 12:12 pm
 
nmm1 at (no spam) cam.ac.uk wrote:
Quote:
In article<n9vfo7-n0j.ln1 at (no spam) ntp.tmsw.no>,
Terje Mathisen<"terje.mathisen at tmsw.no"> wrote:
In fact, my .sig dates from the moment when I suddenly realized that all
programming beyond naive/brute force is a form of caching.

I think I know what you mean, but feel that you are distorting the
term unreasonably. A hell of a lot of things are reasonably viewed
as caching (B-trees, for example), but there are plenty of things
where the solution is to use an equivalent but different mathematical
formulation.

And it does NOT apply to software engineering, which is definitely
part of programming, where the key is to constrain yourself as much
as possible consistent with achieving your objective.

I'm nothing if not inconsistent:

My caching definition (or at least the exact border) can change from day
to day and situation to situation. :-)

Terje

--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"
 
Rob Warnock...
Posted: Wed Oct 13, 2010 3:32 pm
 
Terje Mathisen <"terje.mathisen at tmsw.no"> wrote:
+---------------
| Bakul Shah wrote:
| > Rob Warnock wrote:
| >> Microsecond? *MICROSECOND?!?* Nick, remember, a microsecond is a *MILE*
| >> when it comes to GPS. No, the GPS constellation is synchronized to
| >> within a few *nano*seconds.
| >
| > Rob, how do you map a microsecond to a mile? And a mile where?
| > I am being lazy (& dense) but I couldn't figure out how. Even
| > light travels 300 meters in a microsecond.
|
| I think Rob was talking about 5 us, which is close enough to a mile, at
| least for US engineers (and track & field runners, hereabouts we prefer
| 1500m).
+---------------

Yes, I was indeed thinking of what I'd posted ~15 minutes earlier:

Remember, a 5 microsecond time error is nearly a *mile* position error!

but in the restatement dropped the "5". Mea culpa. (*blush*)

And, yes, I do remember Grace Hopper -- after she showed us the 11.8"-long
bundle of 25-pair-cable "nanoseconds" -- holding up a ~1000'[1] coil of wire
and saying, "...and this is a microsecond".


-Rob

[1] Well, ~984 feet.

-----
Rob Warnock <rpw3 at (no spam) rpw3.org>
627 26th Avenue <URL:http://rpw3.org/>
San Mateo, CA 94403 (650)572-2607
 
MitchAlsup...
Posted: Wed Oct 13, 2010 4:47 pm
 
On Oct 13, 8:17 am, "Ken Hagan" <K.Ha... at (no spam) thermoteknix.com> wrote:
Quote:

Nice story, but light travels rather slower when not in a vacuum, no?

In all normal materials, yes. However plasma's are different.

http://farside.ph.utexas.edu/teaching/em/lectures/node100.html

Mitch
 
Andrew Reilly...
Posted: Wed Oct 13, 2010 4:57 pm
 
On Wed, 13 Oct 2010 08:13:39 +0200, Terje Mathisen wrote:

Quote:
In fact, my .sig dates from the moment when I suddenly realized that all
programming beyond naive/brute force is a form of caching.

And runs into the same (large) problem as hardware caching: invalidation
protocols. Lots and lots of bugs are the result of using a cached value
after it stops being true.

Invalidation is hard to get right in software: one doesn't tend to
include the right amount of transparency or introspectionyness.

Cheers,

--
Andrew
 
Ken Hagan...
Posted: Wed Oct 13, 2010 5:17 pm
 
On Wed, 13 Oct 2010 12:32:29 +0100, Rob Warnock <rpw3 at (no spam) rpw3.org> wrote:

Quote:
And, yes, I do remember Grace Hopper -- after she showed us the
11.8"-long
bundle of 25-pair-cable "nanoseconds" -- holding up a ~1000'[1] coil of
wire
and saying, "...and this is a microsecond".

Nice story, but light travels rather slower when not in a vacuum, no?
 
Andy \"Krazy\" Glew...
Posted: Wed Oct 13, 2010 7:47 pm
 
On 10/12/2010 11:25 PM, Terje Mathisen wrote:
Quote:
nmm1 at (no spam) cam.ac.uk wrote:
In article<nareo7-sdh.ln1 at (no spam) ntp.tmsw.no>,
Terje Mathisen<"terje.mathisen at tmsw.no"> wrote:

Anyway, the speed of light is a pretty good proof of the superiority of
the metric system (300,000 km/s) or even better, the metric A paper
formats: A regular A4 sheet of paper is extremely close to a ns in
length. :-)

Nah. A nanosecond is a foot :-)

1 foot: 30.48 cm
A4 paper: 29.7302 cm (From the definition, usually given as 29.7)

1 ns: 29.9792 cm

I.e. 2.5 mm vs 5.0 mm error.

I rest my case. :-)

Terje

Nick was using the English nanosecond. Smile
 
Bakul Shah...
Posted: Wed Oct 13, 2010 9:06 pm
 
On 10/13/10 4:32 AM, Rob Warnock wrote:
Quote:
Terje Mathisen<"terje.mathisen at tmsw.no"> wrote:
+---------------
| Bakul Shah wrote:
|> Rob Warnock wrote:
|>> Microsecond? *MICROSECOND?!?* Nick, remember, a microsecond is a *MILE*
|>> when it comes to GPS. No, the GPS constellation is synchronized to
|>> within a few *nano*seconds.
|
|> Rob, how do you map a microsecond to a mile? And a mile where?
|> I am being lazy (& dense) but I couldn't figure out how. Even
|> light travels 300 meters in a microsecond.
|
| I think Rob was talking about 5 us, which is close enough to a mile, at
| least for US engineers (and track& field runners, hereabouts we prefer
| 1500m).
+---------------

Yes, I was indeed thinking of what I'd posted ~15 minutes earlier:

Remember, a 5 microsecond time error is nearly a *mile* position error!

but in the restatement dropped the "5". Mea culpa. (*blush*)

And here I was thinking Terje's quip about US engineers doesn't apply
to you:-) Funny in a discussion about relativistic correction on the
order of 10^-10!
 
Paul A. Clayton...
Posted: Sun Oct 17, 2010 7:07 pm
 
On Oct 16, 10:11 pm, MitchAlsup <MitchAl... at (no spam) aol.com> wrote:
[snip lots of interesting information]
Quote:
The concordance problem is not about pipelines, it is about when two
different calculations merge as operands at a new instruction;
deciding when to start processing that new instruction.

My point was that a coprocessor is highly independent,
so a two-way superscalar could be a little more like
two scalar cores (and I thought you implied that a
scalar asynchronous core was significantly less complex).



Paul A. Clayton
just a technophile
 
Jason Riedy...
Posted: Sun Oct 17, 2010 8:50 pm
 
And Andy Glew writes:
Quote:
It's easiest to think of doing it in the absence of pipelined ALUs:
dispatch an operation, wait until complete, dispatch another. Then
clearly we can take advantage of variable latency.

Equivalently, if we dispatch only independent operations down the pipeine.

And somehow massively multithreaded machines like the Cray XMT have
utterly missed this point. Isolate the high variable latency bits in
the memory controller, and let the ALUs rip on instructions from
independent streams.

I don't know their low-level details, but it mystifies me that they're
still at 500MHz (aka 12.5MHz for a single stream).

Jason
 
 
Page 4 of 4    Goto page Previous  1, 2, 3, 4
All times are GMT
The time now is Thu Jul 24, 2014 12:21 pm