Profile cover photo
Profile photo
Barak Ori
Barak's posts

Tomorrow at 5 PM, my band is playing at Levontin 7 in Tel Aviv. I know for many of you there just isn't enough time to buy airline tickets and come to Tel Aviv. For the rest, you can come see me make a fool of myself.
The show (7 bands, each playing 3 songs, starts at 4 PM), is free. The Beer isn't...
(Oh, and don't be fooled by my avatar, I'm not playing the tuba...)

Post has attachment

Post has shared content
ראיתי את זה ב-Post של ג'רי ראיין...
Ok, now THIS is AWESOME.
This just might have to go on my itunes playlist... ;-)
"Seven of Nine" by Future Idiots:

Post has attachment
כן, הצלחתי לסחוב את הקרטון הענקי הזה (גובה 2.5) בתוך הרנו סניק מאיקאה הביתה. עכשיו רק צריך להרכיב את הארון...

Post has shared content
I used AltaVista way back, now I use Google...
I'm saddened to see AltaVista being shut down. +Danny Sullivan wrote a nice article detailing the good and bad parts of AltaVista's history.

I was an observer of some of AltaVista's early days, and this is a story of some of those experiences.  In 1996, I started as a researcher at Digital Equipment Corporation's Western Research Lab (DEC WRL), fresh out of graduate school (where I did work on compiler optimization).  One of the things that attracted me to DEC WRL was that while it was a smallish lab of about 25 researchers, people were working on a wide variety of different kinds of things.  I knew that I wanted to broaden the set of topics that I worked on, and SRC and WRL seemed like ideal places to do this, since they were engaged in a dizzying array of projects in different areas, including operating systems, storage systems, user interfaces, computer science theory, mobile hardware design, etc.

Around this time, AltaVista was getting started as a collaboration between several researchers at DEC WRL and DEC SRC (our sister lab a couple of blocks away), including Mike Burrows and Louis Monier.  When I started at WRL, AltaVista's machines (big, refrigerator-sized DEC Alpha machines) were actually in our lab's machine shop, and over a period of months, their numbers gradually grew to fill many nooks and crannies of the lab.  When using the modest weight set in one corner of the lab, one had to be careful to avoid kicking out the power cord on one of these machines (in total I think there ended up being 17 or 18 of these machines scattered around the machine shop, weight room, etc.).  A few months after I started at WRL, the AltaVista folks moved out to a separate building.

People at SRC and WRL continued to collaborate with AltaVista over the next few years (until AltaVista was sold to CMGI in 1999).  My colleagues +Krishna Bharat,+Andrei Broder, +Monika Henzinger, and Puneet Kumar, along with a SRC intern from that summer
+Suresh Venkatasubramanian built a system called the Connectivity Server ("The Connectivity Server: fast access to
linkage information on the Web",, which essentially was a compressed representation of the entire graph structure of the web pages from a large AltaVista crawl that was held in memory for efficient access.  It had convenient APIs that allowed you to retrieve for a given URL what other URLs that page pointed to, and more importantly, which pages pointed to that page.  I had found an interest in the field of information retrieval, as had many others at SRC and WRL, because it seemed clear that providing high quality search was going to be really important as the web grew.

Using the Connectivity Server, my colleague +Monika Henzinger and I started to look at what you could do with the connectivity structure of the web graph, and we started to tackle the problem of finding related pages.  Given one or more seed URLs (say, we wanted to find algorithms that could come up with other relevant pages (e.g.,, etc.).  We tried a few different algorithms using the connectivity
structure alone, and figured we would need to add other features like the actual text on the pages in order to get things to work well.  Surprisingly, using just the connectivity structure and relatively simple algorithms such as just counting co-citations (e.g. just count how many pages that linked to also linked to, and completely ignoring the text on the page worked surprisingly well.  We wrote up our results in a paper in the WWW'99 conference ("Finding related pages in the World Wide Web",  

The code we had written for our research was pretty close to production-quality code and I had spent a bunch of time doing performance tuning on the code, so that it ran in about 100 ms , which was plenty fast to put into production.  (I usually can't resist trying to make code faster: my colleague +Sanjay Ghemawat has accused me of being a "performance fetishist").  I contacted some folks at AltaVista, and they assigned a person to work with me to get this into production, which I was excited about (it's always great to see people directly using the results of your work).  AltaVista had grown rather rapidly in the period from 1998 to 1999, and I think they were up to several hundred people, and their hiring standards had probably not been the highest during this period.  I packaged up my code and sent it off to the person assigned to work with me, along with detailed instructions about how to compile it, try it out, etc.  After not hearing anything back for a week, I contacted him and asked if he had managed to compile the code and get it to run.  He replied, "I would have, but the file you sent me had this funny 'dot tee aaa rrrr' extension on it, and I didn't know what to do with it."  (A .tar file is a common archive format that every competent software engineer should know how to deal with).  At this point, I decided technology transfer from DEC WRL to AltaVista was not going to go well, and I made a decision to leave the research world and go to the world of startups, which is how I ended up at Google.

When I interviewed at Google a few months later, one of my interviewers was my future officemate +Georges Harik.  At the time he was starting to think about the problem of finding related pages, and so he asked me some questions during our interview session about this.  We had a great discussion during my interview about the tradeoffs of various techniques.

With all that said, I have a fondness in my heart for AltaVista: it was how I first became interested in building large-scale systems for information retrieval for millions of users.

AltaVista, we shall miss thee, now that Yahoo has kicked out your proverbial electrical cord.

Post has attachment

כל המשפחה נוסעת בתחילת יוני לחודש בניו-יורק (עבודה) ועוד קצת חופש.
למישהו יש טיפ לגבי מלון דירות נחמד ב-Upper West (עדיף) או Chelsea? או אולי דירות להשכרה לטווח קצר (חודש)?

Post has attachment

Post has attachment

Post has attachment
תמיד הערתי למי שמצלם וידאו בטלפון לגובה - הנה עוד הסבר:
Vertical Video Syndrome - A PSA
Wait while more posts are being loaded