Supporting cluster computing in the research community

<span class="byline-author">Posted by Christophe Bisciglia, Senior Software Engineer, Google Research and James French, Program Director, National Science Foundation</span><br /><br />Research and development is the foundation of innovation in the technology industry, and both Google and the National Science Foundation (NSF) are committed to making substantial investments in computer science research and education to ensure that our students, faculty and scientists remain on the leading edge of computing<span style="background-color: rgb(255, 255, 255);"> and have the tools necessary to make significant advances</span>. As the technology industry moves into an exciting new phase of massively scaled, highly distributed computing, academic users have expressed a need for resources to engage and explore this emerging model, which is already responsible for many "internet-scale" applications that we now take for granted. That's why we're excited to announce that the NSF will use resources made available via Google and IBM's Academic Cluster Computing Initiative (ACCI) to reach the broader research community and explore new applications for massively scaled, highly distributed computing.<br /><br />In October of 2007, Google and IBM <a href="http://googleblog.blogspot.com/2007/10/let-thousand-servers-bloom.html">announced</a> the first pilot phase of the ACCI, which granted several prominent U.S. universities access to a cluster of thousands of processors running Apache's Hadoop, an open source distributed computing platform inspired by Google's file system and MapReduce programming model. (There's a <a href="http://www.youtube.com/watch?v=UBrDPRlplyo">YouTube video</a> showing how students and faculty at the University of Washington have made use of the cluster.) Encouraged by these results, we sought out ways to extend the reach of this technology, and given its prominence in national research funding, the NSF emerged as a natural partner. For decades, the NSF has engaged the U.S. research community by setting research agendas and encouraging the development and adoption of disruptive technology, and this partnership will encourage a broader range of academic researchers to develop innovative new methods of data analysis using the unique advantages of massively parallel systems architecture - without the massive infrastructure costs usually associated with large-scale computer clusters.<br /><br />Starting this year, the NSF will work to evaluate project proposals from academic researchers across many disciplines and select some of the most interesting and promising projects to receive ACCI computational grants. More details, including information on how to apply for access to these resources via the NSF's Cluster Exploratory (CluE) program, will be posted soon on the <a href="http://www.nsf.gov/">NSF site</a>.

Supporting cluster computing in the research community

Posted by Christophe Bisciglia, Senior Software Engineer, Google Research and James French, Program Director, National Science Foundation Re...

Đọc thêm »

Our secret sauce

<span class="byline-author">Posted by Hal Varian, Chief Economist</span><br /><br />I was recently a guest columnist on the <a href="http://freakonomics.blogs.nytimes.com/2008/02/13/bring-your-questions-for-google-economist-hal-varian/">Freakonomics Blog</a>. There were several interesting questions from the readers, but one was quite striking:<br /><br /><div style="margin-left: 40px;"><span style="font-style: italic;">"How can we explain the fairly entrenched position of Google, even though the differences in search algorithms are now only recognizable at the margins? Is there some hidden network effect that makes it better for all of us to use the same search engine?"</span><br /></div><br />It seems that a lot of people are trying to figure out why Google has done so well. The difficulty is that the typical economic forces at work in many technology businesses that lead to entrenchment don't seem to explain our success. Let's take a look at the usual culprits.<br /><br /><b>Supply side economies of scale.</b> This refers to the fact that a larger business may enjoy a cost advantage. The problem is that though there probably are some scale advantages, they get played out at a reasonably small scale. There are plenty of data centers out there and plenty of people that know how to run them efficiently.<br /><br /><b>Lock-in. </b> The idea here is that when users have a high cost of switching to an alternative provider, they can be charged high prices that reflect the fact that they are effectively locked in to a single provider. But if you look at Google's business, the competition is only a click away. Users can trivially switch search engines. Most of our large customers also advertise on other search engines. And most publishers get their ads from a variety of providers, including their own sales force. So there are very small costs of switching to an alternative search engine for users, advertisers, and publishers.<br /><br /><b>Network effects.</b> This refers to a phenomenon where the amount that people are willing to pay for a service depends on the number of people that have already adopted a service. The classic example is a fax machine: the amount that I am willing to pay for a fax machine depends on how many of my correspondents already have one. But this doesn't fit the Google case either: my decision to use Google is irrelevant to other users. It's true that advertisers want to advertise where there are lot of users but that doesn't affect the amount that they are willing to pay on a <i>per user</i> basis. The value of a user to an advertiser depends on how likely he or she is to buy, not how many users there are. A small website about knitting could be a great place to advertise yarn and could charge rates far higher for such ads than a much larger site.<br /><br />If it isn't economies of scale, lock-in, or network effects, what is it that explains Google's success?<br /><br />The answer, at least in my opinion, is a much older economic concept called "learning by doing" that was first formalized by Nobel Laureate Kenneth Arrow back in 1962. It refers to the widely-observed phenomenon that the longer a company has been doing something, the better it gets at doing it.<br /><br />Google has been searching the web for nearly 10 years, which is far longer than our major competitors. It's not surprising that we've learned a lot about how to do this well. We're constantly experimenting with new algorithms. Those that offer an improvement get rolled into the production version; the others go back to the drawing board for refinement.<br /><br />So I would argue that Google really does have a better product than the competition -- not because we have more or better ingredients, but because we have better recipes. And we are continuously improving those recipes precisely because we know the competition is only a click away. We can't fall back on economies of scale, or switching costs, or network effects, to isolate us from the competition. The only thing we can do is work as hard as we can to keep our search quality better than that of the other engines.

Our secret sauce

Posted by Hal Varian, Chief Economist I was recently a guest columnist on the Freakonomics Blog . There were several interesting questions f...

Đọc thêm »

The art of science and engineering

<span class="byline-author">Posted by Michael Weiss-Malik, Software Engineer</span><br /><br /><span style="font-style: italic;font-size:85%;" >(Cross-posted on the <a href="http://google-latlong.blogspot.com/2008/02/art-of-science-and-engineering.html">Google Lat Long Blog</a>.)</span><br /><br /><a href="http://moon.google.com/" id="x_-6" title="Google Moon">Google Moon</a> and <a href="http://mars.google.com/" id="qvro" title="Google Mars">Google Mars</a> are great examples of products that required much more than pure software engineering to produce. There was quite a bit of science, and even a little bit of artistry, that went into <a href="http://googleblog.blogspot.com/2006/03/mars-attracts.html" id="av9n" title="their">their</a> <a href="http://google-latlong.blogspot.com/2007/09/new-moon.html" id="l_-t" title="creation">creation</a>. They both expose large volumes of imagery and information in simple and accessible designs, and it turns out that I'm not the only one who thinks that they qualify as art in this regard.<br /><br />New York's <a href="http://www.moma.org/" id="h6rf" title="Museum of Modern Art">Museum of Modern Art</a> has honored both products by including them in their exhibition <a href="http://www.moma.org/exhibitions/exhibitions.php?id=5632&ref=calendar" id="qq3y" title="Design and the Elastic Mind">Design and the Elastic Mind</a>, which opens to the public on February 24th. The exhibit showcases objects and systems that pair modern design with innovations in science and engineering in creative ways. Google Moon's Apollo landing <a href="http://www.google.com/moon/#lat=0.656869&lon=23.470613&zoom=19&apollo=a11/10" id="d8y7" title="interactive 360-degree panoramas">panoramas</a> and Google Mars' imagery of the largest <a href="http://www.google.com/mars/#lat=-10.487811&lon=-71.103515&zoom=4&q=valles%20marineris" id="j7yx" title="canyons">canyons</a> and <a href="http://www.google.com/mars/#lat=17.811456&lon=-133.242187&zoom=4&map=visible&q=olympus%20mons" id="gv5:" title="volcanoes">volcanoes</a> in the solar system were intended to do exactly that, by applying Google Maps technology to places that are out of this world.<br /><div><br /></div>We're delighted that MOMA saw fit to include us in their lineup, and recommend the show to anyone living in or visiting the New York area. It runs through May 12th.

The art of science and engineering

Posted by Michael Weiss-Malik, Software Engineer (Cross-posted on the Google Lat Long Blog .) Google Moon and Google Mars are great exampl...

Đọc thêm »

A pilot with the Cleveland Clinic for health information access

<span class="byline-author">Posted by Alan Newberger, Engineering Manager</span><br /><br />I suffer from poor eyesight and intense seasonal allergies, but I'm thankful that health issues occupy just a small portion of my life. Even though I'm rather healthy, I sometimes find myself needing access to accurate health information. I can get a long way by searching for health facts online, but I also need to incorporate what I find with my own history of conditions and treatments. I didn't even realize I <i>had</i> allergies until my early twenties -- for more years than I care to admit, I'd forget that the "cold" I came down with in April was suspiciously similar to the one I had at exactly the same time the year before. I've often been overwhelmed when trying to determine or track a condition, because my personal record of health information is either nonexistent, or it's spread on forms and receipts from (at least) a dozen doctors and five insurance companies.<br /><br />Working as an engineer here on the health team, I've been excited to participate in building tools that will help me and others manage our personal health information more effectively. Many innovators in the healthcare industry have worked hard to make results of doctor visits, prescriptions, tests and procedures available digitally. By using the <a href="http://code.google.com/apis/gdata/">GData protocol</a> already offered in many Google products, and supporting standards-based medical information formats like the Continuity of Care Record (CCR), our health efforts will help you access, store and communicate <i>your</i> health information. Above all, health data will remain yours -- private and confidential. Only you have control over when to share it with family members and health providers.<br /><br />This week, we hit another important milestone. We launched a pilot with a medical institution committed to giving patients access to their own medical records: <a href="http://cms.clevelandclinic.org/body.cfm?id=227&action=detail&ref=815">The Cleveland Clinic</a>. A large academic medical center, Cleveland is one of the first partners to integrate on our platform. Because of their size and reach with patients who already have access to their medical records online, Cleveland has been a great partner for us to test out our data sharing model. Patients participating in the Cleveland pilot give authorization via our AuthSub interface to have their electronic medical records safely and securely imported into a Google account. It's great to see our product getting into the hands of end users, and I look forward to the feedback that the Cleveland patients will provide us. <br /><br />Cleveland is just the first of many healthcare providers that will securely send medical records and information via Google APIs at your request. We've been hard at work collaborating with a number of insurance plans, medical groups, pharmacies and hospitals</span>. While this pilot is open initially to just a few thousand patients, I see it as an important first step to show how Google can help users get access to their medical records and take charge of their health information.

A pilot with the Cleveland Clinic for health information access

Posted by Alan Newberger, Engineering Manager I suffer from poor eyesight and intense seasonal allergies, but I'm thankful that health i...

Đọc thêm »

AdSense for video now in beta

<span class="byline-author">Posted by Shamim Samadi, Product Manager and Ryan Hayward, Product Marketing Manager</span><br /><br />In a very short time, watching videos online has become a common pastime, and the imagination of artists and other content owners continues to fuel this trend. Meanwhile, across the industry, advertisers and video publishers continue to look for the best solution that balances the needs of video fans with the need for video sites to generate the revenue that enables them to continue to be creative as they grow.<br /><br />Enter, stage left: the AdSense for video beta. This approach takes the same non-intrusive InVideo ad format used on YouTube and extends it across video partner sites on the Google Content Network. This enables advertisers to run a single campaign across the largest network of online video content.<br /><br />Unique to AdSense for video are text overlay ads contextually targeted to a combination of signals in your videos and on your site. With these overlay ads, the user's experience is not interrupted; users determine how much they want to interact with the ad.<br /><br />We have two major goals: to make it easier for publishers to monetize video online, and for advertisers to learn how to reach the video community. Towards these ends, today we're also launching a single destination with an overview of various options to expand online video opportunities called <a href="http://www.google.com/ads/videoadsolutions/">Google Video Advertising Solutions</a>. We hope this will be your starting point to understand how to leverage the Google Content Network and YouTube to make the most out of the online video experience. You can also see our overview videos for <a href="http://www.google.com/ads/videoadsolutions/advertiser.html">advertisers</a> and <a href="http://www.google.com/ads/videoadsolutions/publisher.html">publishers</a> -- because isn’t video the best way to experience video?<br /><br />We've been working with a number of video partners (e.g. popular destinations like BobVila.com, eHow, MyDamnChannel, ExpertVillage, PinkBike, TheNewsRoom, and social video aggregators like Revver, blip.tv, and GodTube). And we're also working with several key video platform solution solutions like Brightcove, Yume, Tremor Media, and Eyespot Network, who have plugged our ads into their platforms.<br /><br />There are some criteria that publishers have to meet to participate in AdSense for video, which you can learn about on the <a href="http://adsense.blogspot.com/2008/02/fueling-creativity-in-online-video-with.html">Inside AdSense blog</a>.<br /><br />As with many things we do around here, we’ll keep testing various ad formats, iterate and expand, so keep an eye on us.

AdSense for video now in beta

Posted by Shamim Samadi, Product Manager and Ryan Hayward, Product Marketing Manager In a very short time, watching videos online has become...

Đọc thêm »

Superdelegate layer in Google Earth

<span class="byline-author">Posted by Rick Klau, Strategic Partner Development Manager and Elections Hipster</span><br /><br />I'm a political junkie, and as Super Tuesday neared it seemed likely that the race for Democratic nominee would extend well past that big day at the polls. I started looking around for information about the so-called "<a href="http://en.wikipedia.org/wiki/Superdelegate" target="_blank" title=""Super Delegates"">superdelegates</a>" (the party leaders and elected officials who make up 20% of the vote at the convention) -- and was disappointed to find no single resource that provided details on who they were, which candidate they'd endorsed (if anyone) and where they are located.<br /><br />So I went ahead and registered '<a href="http://www.superdelegates.org/Main_Page" target="_blank" title="superdelegates.org">superdelegates.org</a>', installed <a href="http://www.mediawiki.org/" target="_blank" title="MediaWiki">MediaWiki</a> and started populating the site with the info I could find. Thanks to a terrific extension to MediaWiki (<a href="http://www.mediawiki.org/wiki/Extension:KML_Export" target="_blank" title="KML Export">KML Export</a>, written by Juliano Ravasi), it's possible to map the wiki pages to a <a href="http://www.superdelegates.org/Special:KMLExport" target="_blank" title="Google Earth layer">Google Earth layer</a>, which helps to visualize where these delegates are.<br /><br /><div style="text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEguauV5nfzJezfVBMnphzH_RsO4B9NyrGs75J4AhqtelrKgAdcSHTs4WyOcLJ4K6s2yCmDmSSdKof3rZ_e7thVO6LvmFr2qNz-tc0YTz7P4ZcpEdMQsE4AOsichpXvYf_y8083PR5JGdEA6/s1600-h/superdelegates-google+earth-becerra.PNG" onblur="try {parent.deselectBloggerImageGracefully();} catch(e) {}"><img alt="" border="0" id="BLOGGER_PHOTO_ID_5169138331363619890" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEguauV5nfzJezfVBMnphzH_RsO4B9NyrGs75J4AhqtelrKgAdcSHTs4WyOcLJ4K6s2yCmDmSSdKof3rZ_e7thVO6LvmFr2qNz-tc0YTz7P4ZcpEdMQsE4AOsichpXvYf_y8083PR5JGdEA6/s320/superdelegates-google+earth-becerra.PNG" style="cursor: pointer;" /></a><br /></div><br />Since the <a href="http://www.superdelegates.org/" target="_blank" title="superdelegates site">superdelegates site</a> launched two weeks ago, it has been <a href="http://www.cnn.com/video/#/video/politics/2008/02/12/google.superdelegate.cnn" target="_blank" title="featured on CNN">featured on CNN</a>, and tens of thousands of people have stopped by to learn more about superdelegates, understand the process, and add information. Over 60% of the delegates are now on the map and are associated with their endorsed candidate, with more info coming in every day. It's tremendously gratifying to see a community grow around this timely subject- hope you find it useful!<br /><br /><span style="font-weight: bold; font-style: italic;">Update: </span>Since I posted this entry, there have been a few additional improvements to the <a href="http://www.superdelegates.org/" target="_blank" title="superdelegates site">superdelegates site</a>. <span>Every delegate's wiki page now includes a Google Newsbar with scrolling headlines mentioning the delegate. </span><span>Fellow Googler Bob Rose also created an enhanced <a href="http://www.superdelegates.org/Special:KMLExport" target="_blank" title="KML layer">KML layer</a> so that each delegate's placemarker now includes information and links to their</span><span> home state, position in the DNC, and candidate endorsement (with YouTube videos of speeches if available).</span><span style="font-weight: bold; font-style: italic;"><br /><br /><span style="font-size:85%;"><span style="font-style: italic;">Disclaimer: </span></span></span><span style="font-style: italic;"><span style="font-size:85%;">Since superdelegate endorsements are not binding, we can't ensure that the endorsements listed are indicative of future events. But we're excited that people are engaging in politics online by drawing from a compilation of candidate endorsement lists on <a href="http://www.rollcall.com/politics/endorsements.html" target="_blank" title="candidate endorsement lists on Roll Call"><span style="font-style: italic;">Roll Call</span></a>, <a href="http://www.cnn.com/2008/POLITICS/02/19/dem.delegates/index.html" target="_blank" title="CNN's Election Center">CNN's Election Center</a> and other sources to update the site on a daily basis. Other groups and publications may have different superdelegate counts, including candidates' campaigns, the <span style="font-style: italic;">New York Times,</span> and the AP.</span></span>

Superdelegate layer in Google Earth

Posted by Rick Klau, Strategic Partner Development Manager and Elections Hipster I'm a political junkie, and as Super Tuesday neared it ...

Đọc thêm »

A Valentine Treat from Google

<span class="byline-author">Posted by Scott Giambastiani, Executive Chef</span><br /><br /><object height="355" width="425"><param name="movie" value="http://www.youtube.com/v/0JUDnwPwh9U&rel=1"></param><param name="wmode" value="transparent"></param><embed src="http://www.youtube.com/v/0JUDnwPwh9U&rel=1" type="application/x-shockwave-flash" wmode="transparent" width="425" height="355"></embed></object><br /><br />This recipe pairs well with Roederer Estate, Anderson Valley Brut sparkling wine for a Valentine's day treat for two.<br /><br />Ingredients:<br /><br /><span style="font-style: italic;">Yields about 4 cakes</span><br /><br />1 tbsp. butter (You may substitute olive oil.)<br />4 shallots, minced<br />¼ cup mayonnaise<br />1 fuji apple, peeled and grated on a cheese grater, juice reserved<br />1 tsp. lemon juice, fresh<br />2 tsp. tarragon, minced<br />2 tbsp. parsley, minced<br />1 cup Dungeness crab meat, picked free of shells<br />¼ cup breadcrumbs, finely ground<br />Tabasco sauce<br />kosher salt<br /><br />cooking oil (rice bran, canola, vegetable, etc.)<br />metal heart shaped cookie cutter, large size (About 4-6 oz)<br /><br />Roast Pepper Sauce:<br /><br />8 oz. pimento peppers in the jar (You may substitute with Spanish piquillo peppers,<br />or 3 each red bell peppers roasted over an open flame, peeled and seeded), chopped into 1-inch pieces<br />2 tbsp. sherry vinegar (You may substitute with any vinegar)<br />2 tbsp. water<br />2 tbsp. butter (Optional)<br /><br />Procedure:<br /><br />In a sauté pan, sweat the shallots in butter over medium-low heat until sweet, but with no color forming. This will take about 5 minutes. When sweet, allow the shallots to cool in a medium-sized mixing bowl. Add the mayonnaise, grated apple with juice, lemon juice, herbs, and picked crab to the mixing bowl.<br /><br />With a spatula, gently fold the mixture together, so as not to break up any whole crab meat. This will give the cakes a better texture. Fold the mixture just until evenly mixed. Add enough breadcrumbs to bind the cakes. You might need more or less breadcrumbs than stated,<br />depending on the moisture of the crab. Season the cakes with salt and Tabasco to taste. Using a teaspoon, spoon crab mixture into a metal, heart-shaped cookie cutter. Gently press the crab into the mold. Repeat this process until all cakes are molded. Keep the crab cakes refrigerated until ready for service. Prepare the sauce.<br /><br />To prepare the sauce, place the prepared peppers into a small non-reactive sauce pan with the vinegar and water. Bring contents to a simmer and transfer to a bar blender. Add the butter to the blender, cover and puree the sauce until it is smooth. Season with salt and pepper.<br /><br />To finish the crab cake, heat a sauté pan over medium-high heat, add a teaspoon of cooking oil, and sear the crab cake over medium heat until golden brown. This will take 1-2 minutes. Using a spatula, flip the crab cake over and repeat for 1-2 minutes until the crab cake is golden brown on both sides and warm in the center. Piercing the center of the crab cake with a toothpick and feeling the temperature of the toothpick under your lip will give you a good idea how warm<br />the cake's center is.<br /><br />To plate the crab cake, pour about two heaping spoonfuls of warm pepper sauce over a warmed plate. Place the warm crab cake over the sauce and garnish with baby mixed greens (mâche, upland cress) and/or freshly picked herbs (Italian parsley, chervil, tarragon, chives…). Enjoy!

A Valentine Treat from Google

Posted by Scott Giambastiani, Executive Chef This recipe pairs well with Roederer Estate, Anderson Valley Brut sparkling wine for a Valentin...

Đọc thêm »