[astro] Re: [wcl] FW: corrected PPAM paper draft
David Anderson
davea at ssl.berkeley.edu
Sun Nov 8 23:33:52 EST 2009
If you're running recent server software, you can tell which hosts have GPUs
(this, and the client version number, is stored in host.serial_num).
-- David
Travis Desell wrote:
> Yeah, right now I dont think we have a way to find out how many users
> are using GPUs. I know how many WUs are reported from GPUs but not how
> many users.
>
> --Travis
>
>
> On Nov 8, 2009, at 11:24 PM, Anthony Waters wrote:
>
>> This contains the updated abstract and introduction. I could not find
>> any numbers for how many of our users are using GPUs.
>>
>> Since I have to register for classes at 8:30 am tomorrow I will have
>> some time to make changes in the morning, if need be
>>
>> On Sun, Nov 8, 2009 at 9:37 PM, Boleslaw Szymanski
>> <boleslaw.szymanski at gmail.com <mailto:boleslaw.szymanski at gmail.com>>
>> wrote:
>>> Anthony or Travis,
>>>
>>> attached are modified abstract and introduction, based on zip in Anthony
>>> file. Could you merged it into the current source and generate new pdf?
>>>
>>> On Sun, Nov 8, 2009 at 6:38 PM, Malik Magdon-Ismail
>>> <magdon at cs.rpi.edu <mailto:magdon at cs.rpi.edu>>
>>> wrote:
>>>>
>>>> Hi,
>>>>
>>>> I read through. It looks good. I have a very minor suggestion (unless I
>>>> already missed it). In the acknowledgement we may state the number
>>>> of BOINC
>>>> volunteers and what fraction are GPUs. It could go in the
>>>> acknowledgements
>>>> or somewhere in the intro.
>>>> -Malik
>>>>
>>>> --------------------------------------------------------------------------
>>>> Have Faith in Probability
>>>>
>>>> Malik Magdon-Ismail
>>>> Associate Professor of Computer Science
>>>> Rensselaer Polytechnic Institute (RPI)
>>>>
>>>> Office: Rm 312 Lally mail: Rm 207, Lally Building,
>>>> email: magdon at cs.rpi.edu <mailto:magdon at cs.rpi.edu>
>>>> CS Department, RPI,
>>>> tel: 518-276-4857 110 8th Street,
>>>> fax: 518-276-4033 Troy, NY 12180-3590.
>>>> web: http://www.cs.rpi.edu/~magdon
>>>> key: http://www.cs.rpi.edu/~magdon/MalikMagdon-Ismail.PublicKey.asc
>>>> --------------------------------------------------------------------------
>>>>
>>>> On Sun, 8 Nov 2009, Anthony Waters wrote:
>>>>
>>>>> Updated it to be a X4 940, also changed it from Phemom to a Phenom, I
>>>>> can make changes/updates until about midnight tonight (another 5-6
>>>>> hours). I have class tomorrow morning so I won't be able to change
>>>>> anything tomorrow.
>>>>>
>>>>> On Sun, Nov 8, 2009 at 4:37 PM, Andreas Przystawik
>>>>> <andreas.przystawik at uni-rostock.de
>>>>> <mailto:andreas.przystawik at uni-rostock.de>> wrote:
>>>>>>
>>>>>> Just had a short look and one eror I missed first was the CPU we
>>>>>> compare
>>>>>> to. It is of course a PhenomII X4 940, not a PhenomII X2 as
>>>>>> claimed several
>>>>>> times in the text.
>>>>>> ________________
>>>>>> Andreas Przystawik
>>>>>> Institut für Physik
>>>>>> Universität Rostock
>>>>>> 18051 Rostock
>>>>>>
>>>>>> Tel.: +49 381 498 6815
>>>>>> Fax: +49 381 498 6802
>>>>>> ________________________________________
>>>>>> Von: awaters1 at gmail.com <mailto:awaters1 at gmail.com>
>>>>>> [awaters1 at gmail.com] im Auftrag von Anthony
>>>>>> Waters [watera2 at cs.rpi.edu]
>>>>>> Gesendet: Sonntag, 8. November 2009 19:46
>>>>>> An: astro at cs.rpi.edu <mailto:astro at cs.rpi.edu>
>>>>>> Cc: Andreas Przystawik; Travis Desell; David Anderson
>>>>>> Betreff: Re: [astro] Re: [wcl] FW: corrected PPAM paper draft
>>>>>>
>>>>>> Fixed the spelling of NVIDIA to be Nvidia across the paper
>>>>>>
>>>>>> On Sun, Nov 8, 2009 at 1:41 PM, Anthony Waters <watera2 at cs.rpi.edu
>>>>>> <mailto:watera2 at cs.rpi.edu>>
>>>>>> wrote:
>>>>>>>
>>>>>>> I've made most of the specific comment changes to get things moving,
>>>>>>> considering it is due 11/9 if I remember correctly.
>>>>>>>
>>>>>>> On Sat, Nov 7, 2009 at 3:44 PM, Malik Magdon-Ismail
>>>>>>> <magdon at cs.rpi.edu <mailto:magdon at cs.rpi.edu>>
>>>>>>> wrote:
>>>>>>>>
>>>>>>>> Hi Travis,
>>>>>>>>
>>>>>>>> Can you circulate a new version.
>>>>>>>>
>>>>>>>> -Malik
>>>>>>>>
>>>>>>>>
>>>>>>>> --------------------------------------------------------------------------
>>>>>>>> Have Faith in Probability
>>>>>>>>
>>>>>>>> Malik Magdon-Ismail
>>>>>>>> Associate Professor of Computer Science
>>>>>>>> Rensselaer Polytechnic Institute (RPI)
>>>>>>>>
>>>>>>>> Office: Rm 312 Lally mail: Rm 207, Lally Building,
>>>>>>>> email: magdon at cs.rpi.edu <mailto:magdon at cs.rpi.edu>
>>>>>>>> CS Department, RPI,
>>>>>>>> tel: 518-276-4857 110 8th Street,
>>>>>>>> fax: 518-276-4033 Troy, NY 12180-3590.
>>>>>>>> web: http://www.cs.rpi.edu/~magdon
>>>>>>>> key:
>>>>>>>> http://www.cs.rpi.edu/~magdon/MalikMagdon-Ismail.PublicKey.asc
>>>>>>>>
>>>>>>>> --------------------------------------------------------------------------
>>>>>>>>
>>>>>>>> On Sat, 7 Nov 2009, Carlos Varela wrote:
>>>>>>>>
>>>>>>>>> Travis, Andreas, et al.:
>>>>>>>>>
>>>>>>>>> Here are my comments on the paper at:
>>>>>>>>>
>>>>>>>>> http://139.30.40.11/ppam2009.pdf
>>>>>>>>>
>>>>>>>>> I am sending them rather than editing directly to avoid
>>>>>>>>> collisions...
>>>>>>>>>
>>>>>>>>> GENERAL COMMENTS
>>>>>>>>>
>>>>>>>>> - The abstract should directly talk about the main contributions of
>>>>>>>>> the
>>>>>>>>> paper (mention the four specific GPU optimizations perhaps
>>>>>>>>> categorized at a
>>>>>>>>> high level)... INSERT SUCH A PARAGRAPH BEFORE "A 17 times
>>>>>>>>> speedup..."... and
>>>>>>>>> to keep size constant, perhaps REMOVE "While traditional... As
>>>>>>>>> such,"
>>>>>>>>>
>>>>>>>>> - If feasible, it would be interesting to quantify:
>>>>>>>>> - what percentage of MilkyWay at home volunteers have GPUs,
>>>>>>>>> - what is the respective processing power of the GPUs compared to
>>>>>>>>> the
>>>>>>>>> CPUs in the M at h project,
>>>>>>>>> - what GPU types are most commonly volunteered, how many GPUs (and
>>>>>>>>> how
>>>>>>>>> many CPUs) are typical per volunteer, etc.
>>>>>>>>> - what is the impact of the numerical precision on the
>>>>>>>>> astroinformatics
>>>>>>>>> application?
>>>>>>>>> - e.g., if we use DP we converge in X iterations, if we use SP we
>>>>>>>>> converge in Y iterations (Y=infinity?), etc; indeed a graph with
>>>>>>>>> precision
>>>>>>>>> in the X axis and iterations to convergence in the Y axis could
>>>>>>>>> help
>>>>>>>>> us
>>>>>>>>> precisely quantify when it is better to tradeoff precision for
>>>>>>>>> speedup.
>>>>>>>>>
>>>>>>>>> - Future directions would be interesting to include (further GPU
>>>>>>>>> optimizations, new GPU hardware generations, better GPGPU language
>>>>>>>>> support,
>>>>>>>>> GPU on BOINC, astroinformatics application, etc...)
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> SPECIFIC COMMENTS
>>>>>>>>>
>>>>>>>>> Section 2:
>>>>>>>>>
>>>>>>>>> "Distributed Computing" --> "distributed computing"
>>>>>>>>>
>>>>>>>>> todays --> today's
>>>>>>>>>
>>>>>>>>> "around 60 times difference" --> ??? I think we want to mention
>>>>>>>>> TFlops/GPU ratio and Tflops/CPU ratio directly.
>>>>>>>>>
>>>>>>>>> Nvidia calls his --> Nvidia calls this
>>>>>>>>>
>>>>>>>>> Folding at Home GPUGRID --> Folding at Home, GPUGRID
>>>>>>>>>
>>>>>>>>> Both cases --> Both projects
>>>>>>>>>
>>>>>>>>> Section 3.1.:
>>>>>>>>>
>>>>>>>>> Title: "different" --> "Different"
>>>>>>>>>
>>>>>>>>> reproducability --> reproducibility
>>>>>>>>>
>>>>>>>>> manufactures --> manufacturers
>>>>>>>>>
>>>>>>>>> "Not until the latest...square root operations" --> REWRITE
>>>>>>>>> PARAGRAPH
>>>>>>>>> INTO TWO--IT IS UNCLEAR
>>>>>>>>>
>>>>>>>>> non-coherence --> non-adherence
>>>>>>>>>
>>>>>>>>> This even --> On the contrary, it
>>>>>>>>>
>>>>>>>>> Section 3.2.:
>>>>>>>>>
>>>>>>>>> memory controller --> the memory controller
>>>>>>>>>
>>>>>>>>> und --> and
>>>>>>>>>
>>>>>>>>> "As the calculations... by itself" --> REWRITE PARAGRAPH INTO
>>>>>>>>> TWO---IT IS
>>>>>>>>> UNCLEAR
>>>>>>>>>
>>>>>>>>> amount of threads --> number of threads
>>>>>>>>>
>>>>>>>>> in flight --> technical term: define before use (and use italics)
>>>>>>>>>
>>>>>>>>> neglible --> negligible
>>>>>>>>>
>>>>>>>>> Combine Terms, SQRT, DIV --> Use different font types (and not
>>>>>>>>> capital
>>>>>>>>> letters) for these optimization names
>>>>>>>>>
>>>>>>>>> figure 1 --> Figure 1
>>>>>>>>>
>>>>>>>>> table 1 --> Table 1
>>>>>>>>>
>>>>>>>>> optimal --> optimized ("optimal" is too strong a word)
>>>>>>>>>
>>>>>>>>> require some additional --> requires some additional
>>>>>>>>>
>>>>>>>>> single precision --> single-precision (also apply to
>>>>>>>>> double-precision,
>>>>>>>>> and throughout paper)
>>>>>>>>>
>>>>>>>>> severely compute bound --> significantly compute-bound (or say
>>>>>>>>> "communication to computation ratio is small")
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Section 4:
>>>>>>>>>
>>>>>>>>> their average --> the average
>>>>>>>>>
>>>>>>>>> and c) and --> and c)
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> References:
>>>>>>>>>
>>>>>>>>> 17) "Pracniques" ---??? Is this the right title?
>>>>>>>>> ----
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> One suggestion: CCGRID 2010's papers are due on November 13th---
>>>>>>>>> isn't
>>>>>>>>> this venue a better target (than PPAM) for this work? It may have
>>>>>>>>> much more
>>>>>>>>> visibility in the distributed computing community... just a
>>>>>>>>> thought
>>>>>>>>> to
>>>>>>>>> consider.
>>>>>>>>>
>>>>>>>>> I can make another iteration on the paper (even editing it directly
>>>>>>>>> if
>>>>>>>>> given the editing token) on Monday at 12noon-3pm (NY time).
>>>>>>>>>
>>>>>>>>> Greetings,
>>>>>>>>>
>>>>>>>>> - Carlos.
>>>>>>>>>
>>>>>>>>> ---
>>>>>>>>> Carlos A. Varela <cvarela at cs.rpi.edu
>>>>>>>>> <mailto:cvarela at cs.rpi.edu>>
>>>>>>>>> Associate Professor
>>>>>>>>> Worldwide Computing Laboratory
>>>>>>>>> Department of Computer Science
>>>>>>>>> Rensselaer Polytechnic Institute
>>>>>>>>> 110 8th Street, Troy, NY 12180, USA
>>>>>>>>> http://wcl.cs.rpi.edu/
>>>>>>>>> Ph: +1 (518) 276-6912 Fax: +1 (518) 276-4033 Office: Lally 308
>>>>>>>>> ---
>>>>>>>>>
>>>>>>>>> On Nov 6, 2009, at 7:03 PM, Andreas Przystawik wrote:
>>>>>>>>>
>>>>>>>>>>> Mabye link to the pdf so people can download that easily?
>>>>>>>>>>
>>>>>>>>>> That was in the first email I have sent to everyone (or isn't
>>>>>>>>>> astro at cs.rpi.edu <mailto:astro at cs.rpi.edu> the right mailing
>>>>>>>>>> list?) just before sending the
>>>>>>>>>> zip file
>>>>>>>>>> to you:
>>>>>>>>>>
>>>>>>>>>>> The new version is here: http://139.30.40.11/ppam2009.pdf
>>>>>>>>>>
>>>>>>>>>> Got it lost somehow? But if I look to my webserver log, at
>>>>>>>>>> least one
>>>>>>>>>> person downloaded it already (IP 74.76.53.44) ;)
>>>>>>>>>>
>>>>>>>>>> Andreas
>>>>>>>>>>
>>>>>>>>>> Von: Andreas Przystawik
>>>>>>>>>> Gesendet: Freitag, 6. November 2009 21:19
>>>>>>>>>> An: Travis Desell; astro at cs.rpi.edu <mailto:astro at cs.rpi.edu>;
>>>>>>>>>> David Anderson
>>>>>>>>>> Betreff: corrected PPAM paper draft
>>>>>>>>>> Hello all!
>>>>>>>>>>
>>>>>>>>>> Looks like Travis accidentaly compiled the pdf with an early
>>>>>>>>>> version
>>>>>>>>>> of
>>>>>>>>>> the third section. As he said he's quite busy with the last
>>>>>>>>>> lines of
>>>>>>>>>> his
>>>>>>>>>> thesis and I have still 4 weeks left, I've taken the freedom
>>>>>>>>>> to put
>>>>>>>>>> the
>>>>>>>>>> current version in and also correct a few small errors and
>>>>>>>>>> polished
>>>>>>>>>> some
>>>>>>>>>> rough edges I found so far.
>>>>>>>>>>
>>>>>>>>>> The new version is here: http://139.30.40.11/ppam2009.pdf
>>>>>>>>>>
>>>>>>>>>> My main point of criticism left is that some sentences appear
>>>>>>>>>> thrice
>>>>>>>>>> in
>>>>>>>>>> the text, first in the abstract, then in the introduction and
>>>>>>>>>> finally in the
>>>>>>>>>> summary again.
>>>>>>>>>> ________________
>>>>>>>>>> Andreas Przystawik
>>>>>>>>>> Institut für Physik
>>>>>>>>>> Universität Rostock
>>>>>>>>>> 18051 Rostock
>>>>>>>>>>
>>>>>>>>>> Tel.: +49 381 498 6815
>>>>>>>>>> Fax: +49 381 498 6802
>>>>>>>>>> Von: Travis Desell [deselt at cs.rpi.edu]
>>>>>>>>>> Gesendet: Freitag, 6. November 2009 08:56
>>>>>>>>>> An: astro at cs.rpi.edu <mailto:astro at cs.rpi.edu>; Andreas
>>>>>>>>>> Przystawik; David Anderson
>>>>>>>>>> Betreff: PPAM paper draft
>>>>>>>>>>
>>>>>>>>>> I've attached a draft of the PPAM paper, please read and send
>>>>>>>>>> comments as
>>>>>>>>>> the deadline is november 9th.
>>>>>>>>>>
>>>>>>>>>> You can download it at:
>>>>>>>>>>
>>>>>>>>>> http://www.cs.rpi.edu/~deselt/ppam2009.pdf
>>>>>>>>>>
>>>>>>>>>> --Travis
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> ----------------------------------------------------------------------------------------------------------
>>>>>>>>>> Travis Desell
>>>>>>>>>> <deselt @ cs.rpi.edu <http://cs.rpi.edu>>
>>>>>>>>>> 1-518-867-1054
>>>>>>>>>> Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ )
>>>>>>>>>> MilkyWay at Home ( http://milkyway.cs.rpi.edu/ )
>>>>>>>>>> Rensselaer Polytechnic Institute, 110 8th Street, Troy NY
>>>>>>>>>> 12180, USA
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> ----------------------------------------------------------------------------------------------------------
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>
>>>
>>>
>>> --
>>> Boleslaw Szymanski
>>> Claire and Roland Schmitt Distinguished Professor of Computer Science
>>> RPI, Troy, NY
>>>
>> <main.pdf><ppam2009.zip>
>
> ----------------------------------------------------------------------------------------------------------
> Travis Desell
> <deselt @ cs.rpi.edu <http://cs.rpi.edu/>>
> 1-518-867-1054
> Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ )
> MilkyWay at Home ( http://milkyway.cs.rpi.edu/ )
> Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA
> ----------------------------------------------------------------------------------------------------------
>
>
>
More information about the astro
mailing list