PAPER 11/15: Dynamic speed scaling to manage energy and temperature
5 views
Skip to first unread message
Guofeng
Nov 13, 2007, 11:16:20 PM11/13/07
to ASU:CSE520 FALL 07 Advanced Computer Architecture
Dynamic speed scaling to manage energy and temperature
Bansal, N. Kimbrel, T. Pruhs, K.
T.J. Watson Res. Center, IBM, Yorktown Heights, NY, USA;
This paper appears in: Foundations of Computer Science, 2004.
Proceedings. 45th Annual IEEE Symposium on
Publication Date: 17-19 Oct. 2004
On page(s): 520- 529
ISSN: 0272-5428
ISBN: 0-7695-2228-9
INSPEC Accession Number: 8331118
Digital Object Identifier: 10.1109/FOCS.2004.24
Posted online: 2004-12-13 08:36:26.0
Mike
Nov 20, 2007, 4:46:20 AM11/20/07
to ASU:CSE520 FALL 07 Advanced Computer Architecture
First, this is a very long paper. It discusses previous research in
speed scaling processors in a heavily mathematical focus and then
delivers a new algorithm and compares it with all the algorithm
defined in terms of heat power and speed. While this is an excellent
analysis it does not boil down to a 20 minute presentation while still
delivering the math this paper is focused on and therefore might not
have been an ideal choice for a selection.
Contributions of Paper
The authors "initiate the study" of speed scaling, performing the
first "theoretical investigation of the area."
The authors perform proofs on many existing algorithms that the
original authors did not explicitly provide.
The authors propose a new online algorithm, BKP, and show that "no
deterministic online algorithm can have a better competitive ratio."
Strengths
An at length analysis of existing speed scaling algorithms with
extensive discussion on their strengths weaknesses and other related
costs.
Well supported mathmatical argument with clearly defined assumptions
that the reader can analyze himself.
Weaknesses
The paper does not spend the majority of its time discussing anything
the authors themselves invented. Rather it spends many pages on
analysis of previous works. A seperate paper could have been made to
introduce their new algorithm in a briefer context.
The paper is exclusively theoretical investigation. There are no
experimental results or grounded analysis supported by measured data.
This paper would make a very appropriate chapter of a text book.
speed scaling processors in a heavily mathematical focus and then
delivers a new algorithm and compares it with all the algorithm
defined in terms of heat power and speed. While this is an excellent
analysis it does not boil down to a 20 minute presentation while still
delivering the math this paper is focused on and therefore might not
have been an ideal choice for a selection.
Contributions of Paper
The authors "initiate the study" of speed scaling, performing the
first "theoretical investigation of the area."
The authors perform proofs on many existing algorithms that the
original authors did not explicitly provide.
The authors propose a new online algorithm, BKP, and show that "no
deterministic online algorithm can have a better competitive ratio."
Strengths
An at length analysis of existing speed scaling algorithms with
extensive discussion on their strengths weaknesses and other related
costs.
Well supported mathmatical argument with clearly defined assumptions
that the reader can analyze himself.
Weaknesses
The paper does not spend the majority of its time discussing anything
the authors themselves invented. Rather it spends many pages on
analysis of previous works. A seperate paper could have been made to
introduce their new algorithm in a briefer context.
The paper is exclusively theoretical investigation. There are no
experimental results or grounded analysis supported by measured data.
This paper would make a very appropriate chapter of a text book.
Pradnyesh Gudadhe
Nov 20, 2007, 9:31:08 AM11/20/07
to asucse520-fall-07-advanc...@googlegroups.com
Please find my critique attached with this email.
Thanks.
Regards,
Pradnyesh
--
Please visit my website here:
http://www.geocities.com/paddyinpilani
Thanks.
Regards,
Pradnyesh
--
Please visit my website here:
http://www.geocities.com/paddyinpilani
peyman...@gmail.com
Nov 22, 2007, 4:24:01 PM11/22/07
to ASU:CSE520 FALL 07 Advanced Computer Architecture
The following is the summary of the online discussions on the paper
"Dynamic speed scaling to manage energy and temperature".
*Paper Review*
Published in 2004, this paper introduces a new algorithm, BKP, to
minimize the energy consumption and temperature of the processor by
speed scaling. They also claim that this algorithm has a competitive
ratio better than OA and AVR and that no deterministic algorithm can
perform better.
*Positive Critiques*
All of the critiques agreed that there were strong mathematical
reasonings and proofs throughout the paper including proofs of
previously introduced algorithms. Also due to the in depth discussion
of the other algorithms in this field, this paper makes a great
reading material.
One of the critiques also observed that the peak temperature can be
reduced by reducing instantaneous power and energy consumption can be
reduced by reducing the cumulative power. It is obvious as to why the
second one holds, however I'm not certain if the first one will be
true in the real environment. I presume this claim was made due to the
assumption that the environment has a fixed ambient temperature,
otherwise the accumulative heat would affect the peak temperature.
*Negative Critiques*
What all the critiques agreed on was that the lack of any sort of
experimental result was a major problem of this paper. One critique
also points out that this was necessary since they had even formulated
the problem as an LP problem.
The length of the paper was another problem that was pointed out,
especially since the paper had included many works that had been
previously published so they could have simply gave a reference to
those. However this is one of the characteristics of a journal paper
that is, it should thoroughly discuss all available techniques, their
benefits and their downsides.
One of the critiques also complains about the material covered in the
abstract of the paper claiming it doesn't reflect the work done by the
paper without going into too much detail.
Another critique complains about the over simplification assumed by
the paper, claiming that linear thermal model, fixed ambient
temperature and allowing heat dissipation only through conduction in
the chip, makes the problem too unrealistic. It is also pointed out
that the authors don't justify the fact that their algorithm is
cooling-oblivious when a major problem in the current generation of
microprocessors, is avoiding hot-spots on the chip.
"Dynamic speed scaling to manage energy and temperature".
*Paper Review*
Published in 2004, this paper introduces a new algorithm, BKP, to
minimize the energy consumption and temperature of the processor by
speed scaling. They also claim that this algorithm has a competitive
ratio better than OA and AVR and that no deterministic algorithm can
perform better.
*Positive Critiques*
All of the critiques agreed that there were strong mathematical
reasonings and proofs throughout the paper including proofs of
previously introduced algorithms. Also due to the in depth discussion
of the other algorithms in this field, this paper makes a great
reading material.
One of the critiques also observed that the peak temperature can be
reduced by reducing instantaneous power and energy consumption can be
reduced by reducing the cumulative power. It is obvious as to why the
second one holds, however I'm not certain if the first one will be
true in the real environment. I presume this claim was made due to the
assumption that the environment has a fixed ambient temperature,
otherwise the accumulative heat would affect the peak temperature.
*Negative Critiques*
What all the critiques agreed on was that the lack of any sort of
experimental result was a major problem of this paper. One critique
also points out that this was necessary since they had even formulated
the problem as an LP problem.
The length of the paper was another problem that was pointed out,
especially since the paper had included many works that had been
previously published so they could have simply gave a reference to
those. However this is one of the characteristics of a journal paper
that is, it should thoroughly discuss all available techniques, their
benefits and their downsides.
One of the critiques also complains about the material covered in the
abstract of the paper claiming it doesn't reflect the work done by the
paper without going into too much detail.
Another critique complains about the over simplification assumed by
the paper, claiming that linear thermal model, fixed ambient
temperature and allowing heat dissipation only through conduction in
the chip, makes the problem too unrealistic. It is also pointed out
that the authors don't justify the fact that their algorithm is
cooling-oblivious when a major problem in the current generation of
microprocessors, is avoiding hot-spots on the chip.
Reply all
Reply to author
Forward
0 new messages