Wednesday, October 7, 2015

HPTS trip report (days 0 and 1)

Last week, from Sunday to Tuesday night, I attended the 16th International Workshop on High Performance Transaction Systems (HPTS). HPTS is an unconventional workshop. "Every two years, HPTS brings together a lively and opinionated group of participants to discuss and debate the pressing topics that affect today's systems and their design and implementation, especially where scalability is concerned. The workshop includes position paper presentations, panels, moderated discussions, and significant time for casual interaction. The only publications are slide decks by presenters who choose to post them." HPTS is by invitation only and keeps it under 100 participants. The workshop brings together experts from both industry and academia so they mix and interact. Looking at the program committee, you can see names of entrepreneurs venture capitalists (David Cheriton, Sequoia Capital), large web companies (Google, Facebook, Salesforce, Cloudera), and academics. HPTS is legacy of Jim Gray, and among its regular participants include Mike Stonebraker (Turing Award winner) and C. Mohan (IBM).

HPTS travel and venue

HPTS is always held at the same venue, Asilomar Conference Grounds, Pacific Grove, CA. My flights Sunday morning (JetBlue BUF-JFK, JFK-SFO) were smooth and on time. I even get to do some writing on the JFK-SFO flight. Since Asilomar is not easily accessible from SFO (or San Jose airport for that matter), I had to rent a car. I used the scenic Route 1 for my drive. It was a 3 hour drive. I stopped a couple of times to take pictures. I made it to the Asilomar conference center at 5pm, just enough time to settle before the dinner at 6pm.

The Asilomar conference grounds is just on the edge of the Monterey Bay. It is overlooking the Pacific, and next to white sand dunes (a nature reserve area) and a nice beach. The barracks were ran down and showing their age. There is a dining hall, a separate building where the HPTS participants dined together as a group (breakfast/lunch/dinner). The talks were held in the chapel, another building close to the dining hall. The program was so full that we would go to our rooms only for sleeping, and that too briefly. After dinner, there was social hour the first night, and lightning talks the next 2 nights, all accompanied by beer and wine. And after 9pm, the crowd moved to Director's cottage, a fancy vacation house for chatting till midnight, lubed by whisky, wine, beer (see, the order is different this time). Then breakfast starts at 7:30am, rinse and repeat each day.

HPTS first night

So Sunday evening, at the first dinner, I somehow sat right next to David Cheriton. He is smart and curious. He asked me to explain about my work, but I wasn't able to communicate the motivation for our hybrid clocks work. I suspect this was partly because we don't share the same terminology/background, and partly because I was unprepared to explain the motivation from a database industry perspective. David was interested and persistent and pushed to understand the problem completely, a trait shared by ultra successful people like him. After spending 10 minutes, I was the party to quit, and told David that I hope to clarify these in the lightning talk, which I proceeded to botch up Monday night :-(

There was a meet and greet after the dinner, from 7-9pm, at the chapel. I am actually a shy guy, but I made an effort to meet people. When I saw a crowd of 3 or more people, I joined to listen and participate. I had nice conversations with the Salesforce crew. I asked them a lot of questions and learned a lot.

At 9pm, the group moved to the director's cottage. I was very tired from the JFK-SFO flight and the 3 hour drive, so after hanging around in the director's cottage for 10 minutes, I went to bed. Since I was jetlagged I woke up at 4am, tried to sleep again but woke up at 6 am. I went for a run, for 3.2 miles. It felt good. Sometimes the best way to fight off exhaustion is by exercising.

Monday morning sessions

Pat Helland (Salesforce) opened the workshop. He said that the workshop is a platform for the academicians and practitioners of databases interact and exchanged ideas. He urged participants to make good use of the 30 minute coffee breaks between the sessions. He said that "actually the sessions are there to punctuate the breaks" :-). Phil Bernstein (Microsoft) asked us to refrain from live blogging or tweeting, as some speakers may talk about confidential technology. Speakers who like release their slidedecks after a week to be posted at the HPTS website. Checking back I see 1/3rds of the slides from talks are posted, and almost all the lightning talk slides are posted.

Here are some of the interesting talks from the two sessions (transactions and applications sessions) Monday morning.

The talk "A1 and FARM scalable graph database on top of a transactional memory layer" from Microsoft Research was about a high performance graph database platform enabled by three hadware trends: 1. inexpensive DRAM (currently $8/GB, machines with 128GB, container will hold more than 100TBs), 2. nonvolatile RAM (DRAM + battery + SSD), and 3. fast commodity networks with RDMA becoming available. (Turns out the Microsoft Research group has an SOSP 15 paper describing this system.)

Kyle Kingsburry, a computer safety researcher at Stripe, gave a nice talk on his Jepsen toolkit for blackbox verification of systems by injecting network partitions and testing basic invariants.

"The Quick and the Dead" talk by James Barrese (PayPal) described Paypal technology initiatives for creating a PAAS platform. These were motivated by a need to innovate quickly and to automate everything.

"From Microservices to Teraservices" talk by Adrian Cockcroft (Battery Ventures) described how microservices and containerazation are useful for accelerating innovation by allowing doing daily releases. Adrian defined a microservice as a loosely coupled service oriented architecture with bounded contexts.

Monday afternoon sessions

In the "Data Federations: An Idea Whose Time Has Come Again", Michael Stonebraker (MIT) talked about their bigdawg polystore platform which will be released on github in a couple months.

"From Trash to Treasure" talk by Pat Selinger (Paradata) was about how to clean and integrate dirty customer data with duplicates, missing values, corrupted values, and invalid values.

I really liked Rodrigo Fonseca's (Brown Univ.) talk on causal metadata tracking in distributed systems. 

Eric Grosse (Google Security Team) gave an informative talk titled "Security Lessons from the Front Lines".

Monday evening lighting talks

Monday evening lighting talks slides are available here. The lighting talks are of 5 minute duration. In my talk, I somehow ran out of 3 minutes in the first 2 slides. After I got the "last 2 minute warning" and I rushed through a couple more slides waiving my arms frantically :-) I should have prepared and practiced before the talk. I was upset about how the talk went but several people (including Phil Bernstein at Microsoft Research) showed interest and approached me later to learn more about the hybrid clocks, which made me feel better.

I will write about day 2, newly-minted McArthur Genius Chris Re's talk, and the overall buzz at HPTS in a couple days. I hope more slides will be added to HPTS site by then.

No comments: