CSCE 689-602 Fall 2020
Submit Papers
This is kind of long. Read the whole thing.
We will be reading a number of papers in this class. I would like the
papers to reflect your interests. You will each find four papers on
microarchitectural prediction to submit for consideration in our class.
The following are constraints on the papers you may submit:
- The papers should be clearly related to microarchitectural prediction.
- At least two of them should come from top architectural conferences ISCA, MICRO, HPCA, ASPLOS, PACT, or ICS.
- Two others may also be from top conferences or may come from lesser known venues, but they must be from ACM, IEEE, or USENIX sponsored conferences or journals. You may also choose a highly cited unpublished arXiv paper.
- You may not submit a paper for which you are an author.
- At least two of the papers should have been published within the past 10 years.
The papers should deal with at least one of the following topics:
- Dynamic branch prediction, including:
- Conditional branch prediction
- Indirect branch prediction
- Return address prediction
- Predictor-driven replacement policy, including:
- Replacement of any level of cache e.g. L1, L2, LLC, die-stacked DRAM, etc.
- Must include non-trivial prediction aspect
- Might include other structures e.g. TLB, BTB.
- Prefetching, including:
- Instruction prefetching
- Data prefetching
- Also might include other structures e.g. TLB, BTB.
- The paper must have a non-trivial prediction aspect.
- Value prediction
- Memory dependence prediction
- Impact of prediction on security, including:
- Speculation-driven side-channel attacks (related to branch prediction)
- Cache-base side channels (where exploiting replacement policy or prefetching is key to the attack)
- Using prediction to detect attacks
Other topics that fall within the scope of microarchitectural prediction may be proposed, but first discuss it
with the professor. Most of the papers are likely to be proposed microarchitectural techniques e.g. a new
prefetcher or branch predictor, but there can also be papers analyzing microarchitectural predictors. For example,
a paper on reverse-engineering a microarchitetural predictor would be fun, or a paper characterizing the
predictability of some aspect of program behavior in a way that gives insight into the design of predictors.
What is a good way to find papers? Go to Google Scholar, type in a search
term like "branch prediction" or "prefetching," unclick "include patents," then look at the list of articles
that come up. There will be a lot. Then, for articles that seem interesting and have a lot of citations, click on "cited by" and see all the papers that cited that paper. There will be some irrelevant ones and some interesting ones. Start clicking and reading abstracts.
One of the papers you submit may become the paper you present to the
class. You may indicate your preference for which paper you would like to
present (details about how to do this will come later). Your selection
of a paper to present might require feedback from the professor.
I will take the set of papers selected by all students and assign
approximately ten of them to each of you as readings for which you will
write a review. That is, you will thoroughly read ten papers and write a
review giving a summary (in your own words), strengths and weaknesses,
questions you have about the paper, and a score giving your assessment
of the merit of the papers. Each student will receive a different set
of 10 papers based (hopefully) on your interest in the topics of the
papers. Later in the semester, we will have a "program committee meeting"
where we discuss many of these papers.
The submission of papers must be completed by September 8, 2020.
Procedure for Submitting Papers
By Thursday, August 27, each of you will have been given an account on the
HotCRP website. HotCRP is a software package used to review papers for real
conferences. You will have received an email with a URL to click that will
take you to the submission website for our class. Your accounts will allow you to submit
as "authors" as well as participate as program committee members.
Update your Profile
When you click on the link, click on "Profile" to update your profile. Your
name should be in there already. You can ignore "collaborators and other
affiliations." Under "Topic Interests" please indicate your interest in the various topics
on a scale of "low" to "high." These choices will be used to assign papers for you to read.
Submit Papers
For each paper
you intend to submit, click on "New submission" which will take you to a
form where you can enter the information for the paper. There are a few relevant fields
you should fill in:
- For "Authors" put your email address and name, not the authors of the original paper. For goodness' sake don't put anyone else as author because they will receive an email from the system.
- For "Title" copy the title of the paper exactly as it appears in the paper.
- Ignore the "Submission" field for now.
- For "Abstract" copy the abstract from the paper.
- The "Contacts" field should already contain your email address. Do not add any mor contacts.
- For "URL" enter a URL where the paper may be downloaded. It would be best to give a link to the paper from the ACM or IEEE digital library or arXiv. If necessary, link to the paper from the authors' website or a conference website if you can find the paper there.
- Under "Topics" tick all the topics that are related to the paper. These topics will be used to assign papers for the class to review.
You may edit your submission until midnight on September 8. Once you have
done this for four papers you are done. The software doesn't actually
let you "Submit" the paper without a PDF submission, but I would like to
avoid having you upload actual copies of those papers to the website for
copyright reasons; we should only be downloading the papers from places
that have permission to store them. If you like, you can upload a blank
PDF to complete the submission process.
HotCRP will email you every time you update something, or (later) every
time a review is available for you to read. This can be annoying and if you
prefer not to receive those emails you can untick the relevant boxes under
"Preferences" for your account.