CREU Progress Report: Week of 12/5/17

Work accomplished: This was a busy week. Lloyd and I are both individually working on the mid-year report amidst also studying for our approaching exams (at the end of this week, and early next week). I am also currently in the process of applying to graduate schools, which will also wrap up next week. Our official research group meetings have ended for the semester.

Goals: My goals for this week with regards to the CREU program are to finish writing the mid-year report. I would also like to pass my final exams so that I am able to graduate this semester. :-) 

Outcomes: Overall, this has been a productive week, and a productive semester. Not as much work on strictly research was accomplished due to the nature of this time of year.

CREU Progress Report: Week of 11/28/17

Work Accomplished: This week, I met with Lloyd for a few hours and we discussed the primary takeaways and insights that we made as the result of our research work this semester. We also discussed a new potential modification to our algorithm that I had been thinking about, as well as next steps for our project. Using this information, we outlined what our end of semester report would look like, and made plans to work on the different components of it in our own time. We plan to do the majority of this report this week. I have also been working on grad school applications.

Goals: Our primary goal for this week is to complete the majority of our end of semester report, so that we won't have to worry about it next week (which is the starting point of our finals period).

Outcomes: This has been a productive week in terms of total work accomplished, and we have tangible next steps to complete, which will lead to more productivity over the next week.

CREU Progress Report: Week Of 11/21/17

Work accomplished: I left Cornell this past weekend for Thanksgiving break. I have been able to do some work on my own, but we did not have a group meeting this week in light of the holidays, and I wasn't able to meet with Lloyd to continue to work on the proof.

Goals: My goals for the upcoming week are to meet with Lloyd to try to finish off some of the things that we started prior to the end of the semester.

Outcomes: Due to the holidays, not as much work was accomplished this week as last week. Next week, my schedule will return to what it is normally.

CREU Progress Report: Week of 11/14/17

Work Accomplished: This week, I spent most of my time considering a different aspect of the cache sharing problem that I hadn't taken much time to consider before. During our group meeting last week, we discussed the possibility of making approximations about cache strategy proofness. Previously, I had been operating on the assumption that only isolation guarantees and Pareto efficiency were viable properties to approximate. This new outlook on the problem has interesting implications.

Goals: My goals for this upcoming week are to continue working on proving the caching algorithm, and modifying it as need be. 

Outcomes: The experience I have had so far with working on the algorithmic part of this problem has been valuable. Most of my research experience thus far has been more heavily focused on implementation. I am making progress on this proof, and hope to have something concrete within the next few weeks. 

CREU Progress Report: Week of November 7th, 2017

Work accomplished: This week, I continued to work on the proof of our new shared cache allocation policy. We had a project group meeting at the end of last week, and during this meeting I got a chance to describe some of the algorithmic issues that I was running into, and brainstorm solutions to them with my group members. Through this process, I was able to resolve some of the issues that I was running into. I left the meeting with new material to work with in writing the proof, but when I continued to work on it, more issues came up that I hadn't seen before. I hope to solve these issues this week. 

Goals: As in prior weeks, my primary goals for the next week are to resolve any lingering issues that I am seeing with the proof, so that we have concrete algorithmic results. This has been a nontrivial process. I also plan to take steps to make these blog posts more comprehensive so as to better document our research groups's progress on this project. 

Outcomes: Progress is being made on the algorithmic component of this project. Lloyd has also made progress on the actual baseline implementation of our system, and has started to run experiments. 

CREU Progress Report: Week of 10/30/17

Work accomplished: I had the opportunity to attend SOSP in Shanghai this week. This was the first systems conference I attended. I went to a lot of talks about cutting edge work in systems, and I also got to meet a lot of really interesting people in the systems community. This prevented me from making as much progress on the proof as I would have liked. However, I managed to work on it for a few hours on the plane, but ended up running into issues regarding the formulation of part of the problem. I intend to bring this up during our next research meeting to see what the other members in the group think about the issue, and brainstorm solutions. 

Goals: My primary objective for next week is to resolve the issue that I'm seeing with the formulation of the proof. I think that this will be a doable upon discussion with the rest of my research group.

Outcomes: Since I went to SOSP, I wasn't able to get the outcomes I wanted (namely, to finish the proof) but I think that I will able to do this within the next week, because this is the only obstacle I have remaining. 

 

CREU Progress Report: Week of 10/23/17

Work accomplished: This week, my research group met and discussed the other issues with the proof that I was running into. This time, they were related to ensuring that the max-flow found by our algorithm was created by the maximum number of users. This meant that instead of using the regular max-flow algorithm, we would have to use the min-cost max-flow algorithm. During this meeting, Lloyd also gave updates regarding the implementation of the system. He also made a significant amount of progress since the last time that we had met. 

Goal: Our research goal for the next few weeks is the same as it was in the past week: to finish the implementation of the system, and to finish writing the proof, and address any other issues that arise in the process. 

Outcomes: Similarly to last week, we succeeded in making progress in both the proof and the implementation of our system. This process has allowed us to more deeply understand our solution to the problem. 

CREU Blog: Week of 10/16/17

Work accomplished: This week, my research group met and discussed the different elements of the proof which I worked on, and the implementation of the system, which Lloyd worked on. There were a few components of the algorithm that we came up with that needed to be considered more thoroughly for the proof to work out. More specifically, we had to modify the algorithm to allow for the caching of partial files.

Goal: Our research goal for the next few weeks is to continue working on the full proof and the full implementation of the system. Since I am the one working on the proof, I will need to take the updates that we decided on during our meeting, and incorporate them into the proof.

Outcomes: Developed a deeper understanding of the limitations of our current algorithm both in the process of writing the proof, and in the in-person discussion that we had during our meeting. As an outcome, I have more ideas for how to proceed moving forward. 

CREU Progress: Week of 10/9/17

This week, during our research group meeting, we had a break through of sorts and have decided on the cache allocation policy that we intend to prove and implement. My task for the upcoming week is to prove that the policy provides certain guarantees for an important set of scenarios. 

We have also decided to increase the frequency of our research meetings to facilitate the progress of this project. 

CREU Progress: Week of 9/25

This week, we designed a new caching policy to address the concerns that we had with our prior version. We were successful in approximating one property about this one policy, but decided that being able to approximate this specific property was less exciting of a research direction than were some of our other options. As a result, we will be pursuing a slightly different variation of the initial problem that we were attempting to solve.

CREU Progress: Week of 9/18

This week me and Lloyd continued to attempt to prove properties about the new cache allocation policy that we designed, prior to him leaving to go to the Tapia conference. For the remainder of the week, I continued to work on the proof.  Exams start next week so I was not able to make as much progress as I would have liked. 

CREU Progress: Week of 9/11

Work accomplished: This week, during our group meeting, I gave a presentation about the Dominant Resource Fairness (DRF) paper, in which I analyzed the differences between DRF and the other caching policies which we discussed in prior weeks.

I also met with Lloyd to discuss ideas for cache allocation policies, and we began the process of proving different properties about these policies.

Goals: We intend on continuing down this same path regarding proving properties about our cache allocation policies for the next few weeks.

Outcomes: Developed a stronger understanding of the problem space and the fundamental limitations that cache allocation policies are faced with.

CREU Progress: week of 9/4

Work accomplished: This week, my research group met for the first time. In this meeting, we reviewed the FairRide and Dominant Resource Fairness papers.

Goal: Our research goal for the next few weeks is to delve further into literature regarding different cache sharing protocols, determine what flaws exist in some of the current state-of-the-art protocols, and think further about what steps could be taken to improve upon them. 

Outcomes: Developed a deeper understanding of the fundamental limitations of cache sharing, and how cache sharing differs from other types of resource sharing, since files can be shared between multiple users.