From: Gordon Mohr (firstname.lastname@example.org)
Date: Mon May 08 2000 - 13:47:05 PDT
Yet another distributed-CPU-and-filesystem startup:
This one looks to be hatched from the same nest (and algorithms)
> From the Ether |Bob Metcalfe
> Centrata to create fastest (virtual) computer
> and largest disk on the Net
> BEWARE OF GREEKS bearing gifts. Kevin Kinsella, my old Sigma Nu
> fraternity brother, bore a one-page plan for Centrata -- a proposed
> spin-off from our alma mater, MIT.
> Kinsella, a renowned entrepreneur, was on the verge of a seed
> investment in Centrata. Being a journalist, however, I don't read
> business plans.
> But glancing at Centrata's one-pager, I couldn't help noticing that
> they are pursuing one of my favorite ideas: Centrata plans to make a
> market for unused computing resources on the Internet and thereby
> assemble "the world's fastest computer and the world's largest hard
> So I went to www.centrata.com to read about Shishir Mehrotra and
> David Ratajczak from the MIT Laboratory for Computer Science, where I
> twice worked.
> They are students of David Karger and Larry Rudolph in a group
> conducting research on high-performance and fault-tolerant
> distributed systems.
> This is the group I associate with Inktomi (via the University of
> California at Berkeley) and with Akamai -- two important dot-com
> infrastructure start-ups.
> Next thing, I was on the phone with CEO Mehrotra, a fourth-year
> Centrata is entered in MIT's "$50k" business plan competition, which
> is where I met Akamai just before that company received its first
> round of venture capital funding.
> The first thing I told Mehrotra was that he probably should not be
> talking to the press quite yet.
> He called back to go on the record with what Centrata is planning,
> except for its first big application.
> In short, Centrata is planning to pay "members" to make available on
> the Internet their unused computing resources; for example, excess
> processor cycles and disk storage on your personal computer.
> Centrata will then charge "customers" to use these resources via
> applications running on its software infrastructure.
> Examples of Centrata applications are Web site testing a la Keynote,
> music downloading a la Napster, electronic mail a la Yahoo, and
> Centrata's first big application, which is off the record. Running
> the world's fastest computer and largest disk is several years out.
> So, were you a Centrata member, a designated portion of your disk
> would become part of a RAID (redundant array of inexpensive disks)
> spread over the Internet. Encrypted data from customer applications
> would be stored on your disk.
> You would not be able to read the data, nor would your disk be the
> only place that any data is stored.
> Centrata data would be redundantly located on many member disks
> according to how it is being used by customer applications.
> You would be paid to be a Centrata member. Benefits to customers
> would be lower cost, higher reliability, and better performance for
> their distributed applications.
> I asked Mehrotra about Centrata's algorithms. He said the basic trick
> is to avoid bottlenecks by randomizing access paths.
> This sounds like Ethernet, only Centrata randomizes over space
> instead of time.
> Centrata is not the first to attempt networking unused computing
> resources. I've previously written about MangoSoft
> (www.mangosoft.com), which combines disks on a LAN. Theirs has been a
> rocky road.
> Another example comes from the ongoing search for extraterrestrial
> intelligence (SETI) at setiathome.ssl.berkeley.edu.
> There you will find 2 million computers registered to invest their
> unused resources in the processing of radio signals from space.
> SETI@Home's member computers volunteer to perform small work units in
> a 2-year-long search.
> Each unit is the scanning of about 100 seconds from a small portion
> of a 2MHz spectrum being recorded by a 305-meter radio telescope in
> Arecibo, Puerto Rico.
> Last time I checked, 2,796 computers had reported search results over
> the preceding 24 hours, delivering 13 trillion floating-point
> operations per second.
> But, so far, the truth is out there.
This archive was generated by hypermail 2b29 : Mon May 08 2000 - 13:47:56 PDT