Assignment-2_Paper-2_Sotiria

“Performance Characterization of a Quad Pentium Pro SMP Using OLTP
Workloads”
Adrian
The reviewer summarizes the important ideas and findings of this paper. In this paper real
hardware is used for the experiments. The reviewer points out the benefits and the
limitations of the experimental setup. Maybe the concern expressed in W3 is too
technical. However, for completeness it could have been included in the paper.
Cansu
The reviewer explains the behavior of TPC-C workload and summarizes the important
ideas and findings of this paper. The reviewer emphasizes the contributions of the paper
for the architectural community. I agree with this point because this is one of the first
papers that reported results about the behavior of the commercial workloads.
W2 is indeed a weak point of the paper. Addressing it, would improve the results.
Especially, because only up to 4 processors are used for the scalability experiments.
Danica
The strong points section is well-written and covers all the important contributions of this
paper. The reviewer suggests that it would be interesting to see the same study using the
DSS workload. Indeed this would make the paper even more interesting. As far as W2 is
concerned, it would be helpful if the reviewer could give more details.
Djordje
S1 is maybe the most important contribution of this paper. This is that commercial
workloads and SPEC have different characteristics. This observation changed the way
computer architects build processors.
Farhan
In S2 the reviewer makes the important point that in every experiment only one
parameter is varied while the rest are kept constant. This methodology should be used for
every experimental setup in order to clearly show the benefits of each technique
individually. The reviewer also suggests that the authors could evaluate the DSS
workload to enhance their study.
Ioannis
In S2 the reviewer states that the experimental results presented include both the
Operating System and the Database CPI. Therefore the results represent the overall
performance of the system in contrast to [1] that does not simulate the Operating System.
As possible improvements the reviewer proposes a study about the DSS workload. The
reviewer considers as negative the fact that the authors do not propose techniques to
improve the performance of OLTP.
Manos
The reviewer focuses on the branch predictor’s behavior and memory bus utilization as
the main strong point of the paper. W1 makes the case that graphs are not easy to
understand. Improving the graphs quality would improve the paper layout in general.
Mutaz
The reviewer expresses positive opinion about the methodology and the presentation of
the material. This is true for a seminal paper like this.
Among the negative points the reviewer lists the fact that only a few cache sizes has been
used for the experiments. This is a limitation of the experimental methodology. Indeed it
would be interesting to see results for more cache sizes or even back-of-the-envelope
calculations.
Onur
S1 is probably one of the most interesting results reported in the paper. This is that “The
time spend on data misses is high for OLTP, ~ 50-70%”.
W2 and W3 are interesting points that would enhance the paper if addressed. So, it would
be interesting to (i) evaluate hardware platforms having different characteristics (ii) study
DSS as well.
Pejman
S1 and S2 summarize the most important contributions of this paper. Furthermore, the
reviewer makes the observation that the evaluation is limited to the hardware platform
used. Indeed it would be interesting to see results for hardware platforms having different
characteristics.
Pinar
The strong points section is well-written and covers all the important contributions of this
paper. As far as W1 is concerned, it would be helpful if the reviewer could provide more
comments. The reviewer also makes the point that the size of the database is important
for the memory hierarchy related problems.
Renata
The reviewer focuses on the fact that the authors have considered only the Informix
database. Indeed it would be interesting to consider other database management systems
and present result about them. The strong comments section summarizes the
contributions of the paper.
Stavros
S2 is an important observation that drove the design of processors targeted for the server
market. I agree with W2 that a simulation environment would allow the authors to further
explore the design space. This is the limitation of using real hardware instead of
simulation environment.
References
[1] P. Ranganathan, K. Gharachorloo, S. V. Adve, and L. A. Barroso. Performance of
Database Workloads on Shared-memory Systems with Out-of-order Processors, ASPLOS,
1998.