What a long week! I arrived at Pittsburgh Sunday May 22 and had one of my housemates pick me up from the airport. With the help of one my professor's grad students, Emil Talpes, I was able to find someone subletting their room in an off-campus townhouse over the summer. The house itself is rather old, just like all the other buildings in Pittsburgh, but the cost of living here is so much cheaper than in California, it makes up for it. Both housemates Carol Pai and Jennifer Ng are sooo sweet. The only things I have to complain about Pittsburgh are that there aren't too many places to eat, and the weather isn't as nice. Maybe I'm too spoiled, coming from sunny Berkeley where the streets are packed with cheap eats.
My first day at CMU was delightful. I met Professor Diana Marculescu and was quite charmed by her. She is perhaps one of the kindest professors I have ever met, very easy to talk to, and undoubtedly a well-accomplished and highly-motivated woman. I'm really lucky to be matched with her through this program. I also met Amanda Rainer, another DMP participant like myself, who will be working with Professor Marculescu as well.
After explaining three possible projects we could undertake, Professor Marculescu showed us around the ECE building called Hamerschlag Hall. If I remember correctly, the building is over 100 years old, making it the oldest on campus. It is quite pretty from the outside, and I would put a picture of it on my website but I left my digital camera cord at Fremont so pictures will have to wait. She then treated us to lunch at Union Grill, what a darling! Afterwards, it was decided that I should work on the project involving variability and energy awareness. Unfortunately Emil Talpes, the grad student who wrote a thesis on this topic, was interviewing for some jobs and couldn't meet me until Thursday. Until then, I spent the majority of the week settling in, reading papers on this topic and subthreshold leakage power, and figuring out how to create this website (it's my first one).
When Emil finally showed up, he gave me an overview of the Globally Asynchronous, Locally Synchronous (GALS) implementation, which involves dividing the processor into separate islands with scalable voltage and their own clock signal so as to prevent clock skew across the entire architecture, and to save power. Diana and Emil's paper was mainly on the models and metrics for variability at the microarchitecture level, and the results of simulating these models under different implementations, e.g., GALS, synchronous. My job will be to add a module that accounts for variability in the simulator, and to replace the existing model for subthreshold leakage with a much better model proposed by Zhang, Wason, and Banerjee of UCSB.
The other grad students in Professor Marculescu's group are Philip Stanley-Marbell, who I won't be able to meet until the DAC conference during mid-June, Koushik Niyogi, and Natasa Miskov. Honestly, they all seem kind of shy except for probably Emil, haha. But that's ok. I'm a shy person too.
Week 2: May 31-June 3
Memorial Day weekend, yay! Both my housemates were out-of-town attending weddings (one of the weddings was actually located at Fremont, CA, my hometown), so I did a little exploring on my own. Turns out Pittsburgh has quite a few pizza places, and being the pizza aficionado that I am, I plan to try them ALL. I've also become addicted to the pastries at Panera Bread, a chain bakery that's quite popular on the east coast but doesn't have too many locations in California. I'll try to post some pics of the pastries later, hehe. I also took the bus to the Waterfront, a shopping center located next to the Ohio River. Pretty nice place, but can't really compare to say, Valley Fair Mall of Santa Clara. Lastly, I visited the newly renovated Carnegie Library of Pittsburgh which is about 5 minutes from where I live. I really like the design-- glass walls, and the inside has a modern feel to it. For sure, I plan to spend countless hours there this summer.
Ok, time to talk about important stuff. The first half of the week, I tried to become familiar with the processor simulator, the code behind it, as well as the module that computes power statistics for the simulator. Since all the code is in C, I had to refresh my memory by looking through Professor Dan Garcia's (my former CS prof) lecture notes... Gotta love those pointers. Afterwards, I wrote code that computes the total number of devices, and from that, the number of critical paths per clock domain. Once I have the number of critical paths, I can model the variability using a probability density function from Emil and Diana's paper "Variability and Energy Awareness: A Microarchitecture-Level Perspective." Plotting these values will require the use of Matlab later on. I was quite pleased to find that I will be using concepts from EE126, a probability and random processes course, in my work. Shows that there is major benefit in taking a variety of courses in my undergraduate career.
Later today, I'll be attending a group meeting in which Emil will present his thesis in preparation for a talk he'll be giving at the ISCA conference. Last week, Natasa already presented to us, and since I was unfamiliar with her material on energy bounds for fault-tolerant nanoscale designs, I just gave some advice on presentation style, haha. Hopefully, I'll be able to ask Emil a realllly tough question today since I've read his paper.
Week 3: June 6-10
Last weekend my housemates drove me to the annual art festival in the city. There were several booths displaying paintings, prints, jewelry, or trinkets for sale, as well as a stage where local bands put up some "interesting" performances. Although I found some of the artwork quite creative, the prices were way above my budget, so I spent my money at the food stands instead. Carol got apple dumplings and ice cream, along with a lobster burger which turned out to be quite scrumptious, while Jenn got fried cactus which disappointingly turned out to be like potato chips, and I bought chicken-on-a-stick (I had a big craving for meat that day) and lemonade. Overall, it was a fun experience and a good opportunity to see the city at its liveliest.
Work this week, to be honest, was not as productive as I would have liked. At first I finished coding the variability probability density functions for each of the clock domains. My next task was to find the overall clock speed of the architecture given the fact that overall clock period equals the maximum mean clock period plus the variation due to within-die fluctuations. However, since I only know the probability distribution of the fluctuations, I am not sure as to how an actual value for overall clock speed could be obtained. The only way I can make sense out of this is to view the overall clock speed as a distribution in itself; however I am not sure if that is what Professor Marculescu wants. Communication with her over email has been slow, and Emil is unfamiliar with the probabilistic aspects of this project, so I feel like I am at a standstill. It is a busy week for both my mentors along with Natasa since they are presenting their work at conferences, so hopefully when things have settled down, I can make much better progress.
Speaking of conferences, I will be attending the DAC conference in Anaheim from June 13-17. My plane leaves tonight, which gives me the weekend to enjoy Anaheim on my own. Honestly, I've begun to miss California for its weather, shopping, and superficial people (me being one of them?), haha.
This is off-topic... Pittsburgh has become VERY humid, but that just gives me an excuse to indulge in ice cream. Lucky for me, there are 3 ice cream shops about 5 minutes away from my house-- Coldstone's, Rita's, and Ben&Jerry's. So far I've only tried Rita's (since there are plenty of Coldstone's and B&J's in California), and my god their gelati is YUMMY!! Basically they layer frozen custard, which is like a creamier version of frozen yogurt, on top of your choice of cream ice or Italian ice (I ordered the cookies and cream ice), which is on top of another layer of frozen custard. I can't believe I'm saying this but I like their gelati more than the ice cream at Coldstone's! Don't you envy me now?
Picture time! Here is Hamerschlag hall, the building I work in:
Here are the pastries at Panera Bread:
And here's me in the city:
Week 4: June 13-17
Didn't think I would say this but it actually feels good to be back in Pittsburgh. Nevertheless, my trip to Anaheim was a welcome break from sitting in front of the computer day after day.
Like I said, the Design Automation Conference didn't start until Monday the 13th so I had the weekend to myself. Two things I absolutely must do each time I visit Southern California: 1) shop for clothes, and 2) eat Chinese food. Thus I devoted one day of the weekend to each of those things, hehe. South Coast Plaza was a bustling shopper's haven as usual. I had surprisingly little energy that day, so I didn't enter my feverish shopping mode like I usually do. What actually ended up boosting my spirits was none other than the happiest place on earth, Disneyland. I didn't enter the park itself but just listening to the Disney background music and enjoying the nightlife in Downtown Disney brought back fond memories of my childhood. Sunday was "Chinese shopping center day." I managed to buy something to eat from each of the three Chinese shopping centers around Colima Rd. in Rowland Heights. By the end of the day, I felt like an overly-stuffed pig.
DAC is the first conference I'd ever been to, and I have to say it was quite an overwhelming experience. Never before had I seen so many people from all around the world come together like this. It turned out that Amanda and I were two of the four undergraduates that received a grant to attend this conference, so we were probably amongst the youngest ones there. Most were either Ph.D students, professors, or people from industry (a substantial few were from Synopsis). The conference was held at the Anaheim Convention center, a 3-story building with large glass walls, high ceilings, and insanely strong air conditioning. The entire bottom floor was designated for company exhibits which are similar to career fairs in that they give away lots of free junk. Quite a few booths were raffling away ipods; of course, I didn't win one and didn't come across anyone who did, so I imagine the odds were pretty lousy. On the other hand, I did receive a T-shirt, wristband, baseball cap (things that I would never be caught dead wearing in public), numerous bags, and a pack of chocolate mints, all of which I threw away by the end of the week (except for the chocolate mints of course and the cap which I'm saving for my shameless father, haha).
Although it sounds like I wasted a lot of time at the exhibits, believe me, the majority of the time I was sitting through sessions held on the second and third floors. I attended two workshops on Monday: "Introduction to EDA for Non-Technical Audience" and "Women in Design Automation." The first workshop was just an overview of the chip industry, how chips are made, etc. Although it was a little too basic for me, I thought it was a nice introduction to the conference overall. The second workshop explored diversity issues, such as how to cope with a culturally diverse environment at the workplace, how to increase diversity in the engineering major and industry, and why there is a lack of women and minorities in the field. Jane Margolis of UCLA was the keynote speaker for this workshop, and in my opinion, she did an excellent job of touching upon all these issues. A couple observations she made through surveying both female and male engineer majors that I found quite interesting were that women tend to be bothered by not fitting into the engineering stereotype-- sitting in front of the computer 24/7 with no life outside of engineering-- more so than men, and also females had very little hands-on engineering experience before choosing their college major compared to their male counterparts. Ms. Margolis suggested that this may trace back to the way parents tend to raise girls and boys differently; ie. males are given more opportunities to develop an interest in engineering as a child. Personally, I never felt like I was given fewer opportunities to develop an interest in engineering than say, my brother. Sure I played with dolls more than I did with legos, but that was because I liked dolls more than legos, not because my parents pushed it on me. Honestly, I was more set on becoming a doctor all the way up until junior year of high school when I realized that it would be a sad waste of math and science talent if I didn't become an engineer. Anyway, the workshop made me realize just how thankful I am to have parents that didn't expect any less of me because of my gender.
Monday evening, my professor and her husband Professor Radu Marculescu, treated all their students to a dinner at Jazz Kitchen in Downtown Disney. That was so much fun, getting to know the grad students and profs better. The profs also brought along their kids; they were so adorable! The food was delicious, Louisiana cajun style. I ordered gumbo, grilled salmon, and Diana insisted that we all have beignets for dessert which are French donuts dipped in powdered sugar. Pure heaven!
Tuesday started with the keynote speech "How Does One Define 'Technology' Now That Classical Scaling Is Dead (and Has Been for Years)" given by Vice President and Chief Technologist of IBM, Bernard S. Meyerson. To sum up his speech, he informed us that blind scaling is a thing of the past, the GHz metric should be replaced by application specific metrics, and application specific computing will keep costs down. Why that is the case I missed because I zoned out somewhere in the middle of his speech, sorry. After opening session, I attended three sessions, each with 3-4 speakers, usually Ph.D students or professors presenting their research. I never had the nerve to go up to the mike and ask a question because a lot of the time, their research was beyond my level, and I had perfect faith that they knew what they were doing anyway. Honestly I think a lot of people go up to the mike to show off that they understood the talk and to brag about their affiliation, haha. The talks I liked most were the simplest ones, when the speaker didn't try to cram every detail about their research into the 30-minute time slot, and tried to make the presentation as high-level as possible. For instance, the talk given by Peter Petrov from U. of Maryland about minimizing the number of bits for the tag ID in Virtual Memory was a very nice, simple talk. People should give more talks like Peter Petrov's. I also attended a panel which is where they gather 8-10 speakers to present their work briefly, then engage in debate for 30 minutes or so. Sadly, this panel called "DFM Rules" was rather lifeless. The chairman would ask a question to the panel, and after about 30 seconds of silence, a speaker would grudgingly give an answer, after which no one would bother to offer a second opinion. Unsurprisingly, I left that session early. Finally, I went to the Ph.D forum where Ph.D students presented their posters, and then attended a banquet for CMU alumni and students. The food was beyond my expectations; I had scallops and two helpings of salmon. I know, salmon for 2 days in a row, but I absolutely love seafood.
Wednesday, or "Wireless Wednesday" as DAC called it, featured a few sessions on wireless technology. Out of the three sessions I attended that day, I attended one on wireless technology called "Wireless Emerging Directions," and I found it to be quite refreshing actually. The first presentation was on cognitive radio, which is a radio that can change transmitter parameters based on interactions with the environment. The second was on multiple input multiple output (MIMO) technology, ie. the use of multiple parallel radios to improve data rate, and the use of spatial multiplexing where the same frequency spectrum is used. The third was on low power wireless systems and how the problems encountered in shorter-range transmission are more on the circuit level, and how energy more than ever needs to be taken into account in addition to bps. I understood these talks much better than the other sessions, partially because of the communications course I took last semester, and also because the speakers were more fluent for whatever reason.
Wednesday night I attended the 42nd DAC party featuring The Fab Four, a band devoted to imitating the Beatles. I forced myself to leave early because I found myself losing all self restraint at the dessert table. The setting was nice-- candlelit banquet room inside the Anaheim Hilton, and the music was decent. I think I would have enjoyed the party a lot more if I hadn't pigged out.
Thursday was my last day at the conference. I attended another keynote address "Innovation in the EDA Business Need not be an Oxymoron", this time given by Ronald A. Rohrer of Cadence Design Systems (He has also taught at UC Berkeley a number of times and was a Berkeley student himself). I felt that the speech was intended for business people specifically, as he stressed the importance of building partnerships, how to define the best product possible, and to have patience in nurturing the product into its final form. Thus, I didn't find the speech quite as interesting, but he gets points for coming from Berkeley. As usual, I attended more talks, and another panel discussion called "DFM and Variability." This time, the panel was much more interesting, the audience had more questions, and the chairman had an excellent sense of humor. One of the big issues they addressed was whether or not taking into account parameter variability would amount to profits in industry. The anticlimactic answer was that it still isn't certain, at least not until a unanimous metric for variability is devised. Of all the sessions I attended, I think I liked this one the most.
Technically, DAC ended on Friday the 17th but Friday was just all-day tutorials, none of which I signed up for. Since my flight back to Pittsburgh was late Friday night, I had another day to relax. Let's see-- I watched Batman Begins which was so-so (I prefer movies with more color than just... black), bought the latest Backstreet Boys CD (it's still growing on me), read some more of Angels and Demons by Dan Brown (I highly recommend that book, by the way, especially if you've been to Rome), and had the best steak in my life at Outback. Quite a nice way to end my week at Anaheim!
Week 5: June 20-24
Since my last entry was long and this week wasn't quite as exciting as last week, I'll make this relatively short. I'm happy to say that I've made considerable progress in my work. Before I go into detail about what I've done, let me explain what variability is (sorry I should have done this a lot earlier). Because of techniques for increasing chip performance and lowering power consumption, the number of critical paths increases which then introduces variation in parameters such as clock speed, temperature, and leakage. These variations decrease the reliability of logic and memory available on chip as well as overall performance. Although variations can be either dynamic or random, my research deals only with the random nature of variability. For instance, if we take into account variability for clock speed, then instead of just having a constant clock speed under all circumstances, we actually have a clock speed that takes on a random value with a Gaussian-like distribution. Hope I didn't confuse you even more.
After finetuning my code, I got it to output the cumulative probability distribution for clock speed variability for each of the 5 clock domains in the GALS processor-- Fetch/Decode, Rename/Dispatch, Integer, Floating Point, and Memory. By seeing which variability value corresponds to a cumulative probability of .5, I can obtain the mean variability for that particular domain. I then added these values to the GALS simulator's clock periods so that more accurate clock periods are used, and ran simulations to obtain performance statistics. My next step will be to compare these statistics to those of the synchronous processor simulator, but Emil is still performing tests on that simulator so I'll have to wait. No worries, I'm a week ahead of schedule, yay!
By the way, if any of you ever decide to come to Pittsburgh, you should visit Walnut St in the Shadyside district. In addition to a few cute shops, they have this quaint creperie called Crepes Parissienes. Normally I'm not really into crepes because so far the ones I've had are either way too filling or not filling enough, but this place gets it just right, and they have tons of flavors, too! (Jenn always gets the french toast crepe, although I thought the strawberry crepe was very good too) Unfortunately they seem to have inconvenient hours-- I visited the place this Thursday evening and the sign said they would be closed until next Tuesday, how random is that??
I'd also like to wish my brother, Kevin, a wonderful birthday. Happy 15th birthday! Thanks for still making me laugh so hard!