About Mentor Project Journal Final Report
| Week 1 | Week 2 | Week3 | Week 4 | Week 5 | Week 6 | Week 7 | Week 8 | Week 9 | Week 10 |
June 14 - 18

Stats and Graphs!

I did manage to fix the bug with the scratchpad memory! It turned out that each scratchpad has its own address offset in memory. I had all of the scratchpads for each processor starting at the same address, which was not very helpful!

Then, Cesare told me that I should edit the stats file to include some information about writes to the scratchpad memory (the log)and run some simulations to make some graphs for the meeting tomorrow. The stats file is about 4000 lines of code, so this task was a little daunting! About 3 hours (and some stress and worry that I wouldn't be able to get everything working on time), I had some write stats printing out that were accurate (yay!). That night, I made graphs for the meeting on Tuesday.

Tuesday's meeting was the first meeting where I actually had something to present. I've just been listening in during the past two, because I was still learning about the project. It was really cool to be able to talk about what I had been doing for the past week at the meeting, and explain the graphs and the work.

A Better Log

The graphs that I made showed that the log generated an overhead in energy, cycles, and Energy Delay Product (EDP). I suggested that this might be due to the fact that the log had to be 4KB to prevent an overflow from occurring. Tali Moreshet, a professor in Engineering at Swathmore who is also working on the project, suggested that data for a specific address only be written to the log once to save space. After the meeting, I made some graphs of the distribution of writes to a specific address for a few benchmarks. It seemed like the change in log type was definitely worth it, so I began to work on that.

Since Cesare was leaving for the rest of the week to go to a conference, we talked on Tuesday afternoon about making a direct cache that would store the addresses from the log. He pointed me in the direction of some files that I should look through to see the methods for the direct cache and how one should be implemented. I was a little nervous, because I haven't gone a day without asking Cesare at least one question! Everything turned out to be fine, though! I was really surprised when I finished implementing the direct cache on Thursday, and it worked! On Friday, I ran some simulations. Now, I just need to make graphs of the differences in writes, overflows, energy, etc. From a brief look, it looks like this change to the log has made a difference - its very exciting to see!


Weekend

Dinner and the Beach!

This weekend in Providence was really hot! On Friday night, I went to Julie's (my roommate) friend's house for dinner with Julie and Kristin. We had a really nice dinner in her backyard, with a dessert of fresh strawberries. It was delicious and really fun!

On Saturday, Kristin decided that it would be a really good idea to go the beach (and I completely agreed!) Kristin, Michelle (also a first year grad student at Brown), and I went to a beach in Narragansett. The weather was perfect! The beach itself was also really, really pretty! We had a blast reading on the beach and playing in the waves. Unforunately none of us managed to escape without being a little sunburned, but it was definitely worth it. On the way home, we decided to grab ice cream for a very, very late lunch. On Saturday night, one of the post docs in Kristin's lab and Julie cooked a mexican dinner. It was delicious! Overall, this has been an amazing weekend.