Greg Linden

"When Amazon got its start, we had brilliant engineers. I didn't know all of them, but I knew some of them. Examples? Shel Kaphan. Brilliant. Greg Linden. Brilliant." - Steve Yegge Shel Kaphan: http://www.geekwire.com/2011/meet-shel-kaphan-amazoncom-employee-1 Greg Linden: http://glinden.blogspot.com/

While often ignored, item-to-item similarity recommenders work really well for new items and have been the basis for Amazon's and YouTube's early recommendation algorithms https://glinden.blogspot.com/2011/02/youtube-uses-amazons-recommendati…

http://glinden.blogspot.com/ Mix of interesting technical articles (most related to search and information retrieval) and tech industry prognostication.

If you are in this, maybe you could find interesting this review of a paper from googlers to approach a CAPTCHA design, in which humans are asked to select the right image rotation: http://glinden.blogspot.com/2009/05/exploiting-spammers-to-make-comput… As always, one of the most interesting part of truly great CAPTCHA systems is that they are advancing the state of the art in image recognition. But on the other hand we still have scams like this, and no real solutions.

I really liked Greg's commentary on RMSE -- summed up my thoughts better than I could: http://glinden.blogspot.com/2009/03/what-is-good-recommendation-algori…

Some of the blogs I read: http://www.scottaaronson.com/blog/ http://weblog.fortnow.com/ http://glinden.blogspot.com/ http://community.livejournal.com/evan_tech/ I used to have a blog called CleverCS that you may have heard of.

Apple advocates it but don't do it themselves. They follow the genius design method. http://www.businessweek.com/magazine/content/05_12/b3925608.htm This article doesn't cite it but hint's at it. Which is obvious for a couple of reasons. Apple is paranoid that anything regarding what they work on get's out. and Apple use the Genius design method. Apple do new things not me to products. Other than that you must take my word for it and those I know at Apple. But again if you can find anything that proves that Apple do usability testing then by all means please share. With regards to Amazon. I know that what matters at amazon is A/B testing. http://glinden.blogspot.com/2007/06/ab-testing-at-amazon-and-microsoft… And looking at their checkout process it doesn't look like anything that have been anywhere near a usability test. But you are missing the point here. The companies didn't become successful because of usability testing that is the crux of the matter and the claim behind the 5USD that is wrong. To the extent it testing your product matters you can ask ANYONE. If it is that bad then why spend 5USD on a beer. But most likely it wont be totally unusable and then the actual evidence of usability test having any positive influence on your likely success.

How is increased FCP, LCP, ttvc better ux and DX? We've had hard data that decreasing page load time make for better user experience[0]. Please don't make this an Us vs. Them thing. [0]: http://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html?m=1

It doesn’t seem intuitive at first but this is a well researched phenomenon. https://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html?m=1

Meaningful improvement is in the eye of the beholder. There's the old tale of how Google found out 500ms in latency was a 1% drop in traffic. http://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html 20 to 17 minutes is a 15% time savings still.

Google's index has been in memory for most of its life now: http://glinden.blogspot.com/2009/02/jeff-dean-keynote-at-wsdm-2009.htm… It's actually more complicated than just a single static index, which is also why it's unrealistic to expect a search engine to be deterministic at scale.

I'm not sure of what it's supposed to prove, who knows how many clients and how much money Steam lost because of slow loading/painting times or just plain broken JS. From experience I can tell that I gave up buying a game on Steam at least once because of broken jQuery. There are real metrics from top industry leaders[1][2] to prove that loading time has a clear impact on conversion rates. >Consider this when you need 5000 npm modules and a team of 10 to compile your startup's login form. Unminified JavaScripts files and bloated JS are both wrong, once again: not sure of what's the point of comparing both. [1] https://medium.com/@vikigreen/impact-of-slow-page-load-time-on-website… [2] http://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html

1999 Amazon when after Ebay with Amazon Auctions, about the same time Amazon zShops (now Marketplace) started. Back then a big bet in terms of money and resources that failed. https://glinden.blogspot.com/2006/04/early-amazon-auctions.html

Really ? https://blog.gigaspaces.com/amazon-found-every-100ms-of-latency-cost-t… http://glinden.blogspot.com.au/2006/11/marissa-mayer-at-web-20.html I'm sure I read essentially the same from ebay as well at some point. It's worse than being a competitive advantage : extra milliseconds seem to matter a lot for the size of the market in the first place.

Performance is directly related to profit on consumer apps. See http://glinden.blogspot.co.uk/2006/11/marissa-mayer-at-web-20.html?m=1 plus like you say PHP might be faster to get started with but becomes much slower on larger projects due to inconsistent error handling and language design.

Google found a 0.5 second delay caused a 20% decrease in repeat traffic, that persisted after the delay went away (http://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html) Amazon found every 100ms slower the site loaded, they lost 1% in revenue: http://www.gduchamp.com/media/StanfordDataMining.2006-11-28.pdf Walmart.com found a very large change in conversion rate based on page load times: ( http://www.slideshare.net/devonauerswald/walmart-pagespeedslide slide 37) My own customers have seen 4x different conversion rate, when grouped by page load time. (i.e. same slide as walmart page 37 above, where the peak is 4x higher than the lowest performing group). Your business probably won't fail because it's slow, but it will certainly make less money because it's slow.

Well, it sure worked for Amazon [1][2]. [1] http://glinden.blogspot.com/2006/01/early-amazon-door-desks.html [2] http://glog.glennf.com/blog/2011/10/16/the_true_story_of_the_amazon_do…

> We decided to build instead of buy Meaning they decided to procrastinate instead of working on their product/service. > we instead chose to purchase solid core doors from Home Depot (~$40 each) Meaning they instead chose to spend more on something that wasn't designed to be a desk than an actual desk. From [1]: Ergonomically, door desks leave a lot to be desired. Keyboards were usually too high. Typing for hours could be uncomfortable. And those angle brackets have sharp edges; accidentally scrapping exposed flesh against those was a mistake that wouldn't be repeated. [1] - http://glinden.blogspot.co.uk/2006/01/early-amazon-door-desks.html

A really good case study on this is Greg Linden and his former startup, Findory. (http://glinden.blogspot.co.uk/2008/01/brief-history-of-findory.html) My personal opinion is that news connects with a core human trait; the need to discover, search for new things. Customizing news is the opposite of this when you're presented with the same thing over and over again.

I have no experience with Ruby template engines, including Crustache and Mustache, but of course it matters about serving clients 40x times faster. 800ms to serve me what could be had in 20ms? Amazon, Google [1][2], and others have talked extensively on speed-to-conversion ratios. In Python land, there have huge improvements in template engines [3]. And they are not done; Jinja(2) author, Armin Ronacher, is working on a Google Summer of Code (GSoC) project [4] to speed up Django and Jinja2 templates. [1] http://googleresearch.blogspot.com/2009/06/speed-matters.html [2] http://glinden.blogspot.com/2006/11/marissa-mayer-at-web-20.html [3] http://stackoverflow.com/questions/1324238/what-is-the-fastest-templat… [4] https://www.djangoproject.com/weblog/2011/apr/25/gsoc/

An alternate perspective is available at http://glinden.blogspot.com/2011/05/eli-pariser-is-wrong.html

My company barely existed in the 90s - it was some research project at Stanford. And feature sets like "processes transactions" absolutely do go away because technology (and its applications) moves forwards. This was a large part of the success of MySQL: relational databases started getting used in applications where transactional integrity really wasn't that important, and they were better off working around it (or just showing inconsistent data to the user) rather than paying the price for transactions. Or look at this recent discussion surrounding Google Megastore: http://glinden.blogspot.com/2011/02/comparing-google-megastore.html Megastore suffers a large latency penalty for ACID semantics. In many applications, it's better not to take that latency hit and instead settle for eventual consistency instead.

Quick view: http://docs.google.com/viewer?url=http%3A%2F%2Fsergey.melnix.com%2Fpub… Abstract: Dremel is a scalable, interactive ad-hoc query system for analysis of read-only nested data. By combining multi-level execution trees and columnar data layout, it is capable of running aggregation queries over trillion-row tables in seconds. The system scales to thousands of CPUs and petabytes of data, and has thousands of users at Google. In this paper, we describe the architecture and implementation of Dremel, and explain how it complements MapReduce-based computing. We present a novel columnar storage representation for nested records and discuss experiments on few-thousand node instances of the system. Overview by Greg Linden: http://glinden.blogspot.com/2010/12/papers-on-specialized-databases-at…

Greg Linden has another take on this, agreeing: "Groupon is not Googly" http://glinden.blogspot.com/2010/12/groupon-is-not-googly.html

Brief summary at http://glinden.blogspot.com/2010/11/update-on-googles-infrastructure.h… Longer summary at http://doubleclix.wordpress.com/2010/11/11/google-a-study-in-scalabili…

Perhaps the interviewer wanted to see if you could think up something like http://glinden.blogspot.com/2006/01/early-amazon-door-desks.html - and in the process test your willingness to be in an environment with such jury-rigged solutions.

Greg posted a nice compilation of shady tactics used during the startup period of major internet companies at http://glinden.blogspot.com/2006/11/ruthless-enough-for-startup.html A quick summary: Facebook: Spammed Harvard Students BitTorrent: Free Porn as Initial Bait MySpace: Spammed >100M on Launch Date It appears as if (slightly) shady techniques can give you the advantage you need to succeed.

From http://glinden.blogspot.com/2008/05/yahoo-builds-two-petabyte-postgres… : Yahoo builds two petabyte PostgreSQL database James Hamilton writes about Yahoo's "over 2 petabyte repository of user click stream and context data with an update rate for 24 billion events per day". It apparently is built on top of a modified version of PostgreSQL and runs on about 1k machines. In his post, James speculates on the details of the internals. Very interesting.

Greg Linden's take on the same: http://glinden.blogspot.com/2010/02/personalization-and-differential.h…

Google's search quality very likely doesn't have very much to do with the PageRank algorithm as such. Certainly PageRank isn't much of a qualitative improvement over prior work like HITS, and any reasonable web-scale search engine can easily incorporate some variant on the HITS/PageRank idea. http://glinden.blogspot.com/2007/09/hits-pagerank-and-keeping-it-simpl…