Forgot your password?
typodupeerror
The Internet

NSF Grants for Decentralized Infrastructure Research 71

Posted by michael
from the routing-around-trouble dept.
billbaggins writes "The NSF has given a grant to the IRIS project to research something called Distributed Hash Tables as a tool for creating networks that don't have "centralized points of vulnerability". The chief purpose seems to be to stop DoS attacks, intentional or otherwise. Check out their press release (text or Word format) and also the news coverage (CNN and NYTimes, among others)."
This discussion has been archived. No new comments can be posted.

NSF Grants for Decentralized Infrastructure Research

Comments Filter:
  • REPEAT (Score:2, Informative)

    by Anonymous Coward
    http://slashdot.org/article.pl?sid=02/09/25/171322 4&mode=thread&tid=95
  • I guess they're more worried about Votester [slashdot.org] than we thought!
  • I smell lawsuit (Score:4, Insightful)

    by Crazieeman (610662) on Sunday September 29, 2002 @02:52AM (#4352715) Journal
    The RIAA won't stand for this. They want the power to DoS and are pushing legislation to make it legal for them. So this will probably be in violation of the DMCA somehow. They'll get their bloodsuc- I mean lawyers right on it.
    • "The RIAA won't stand for this."

      W.T.F.?!? They're trying to create a more robust, secure network. Like, y'know, the internet itself. Or like the phone system. Or like the power grid. Or like the sewers. "No central point of failure"

      What's that got to do with RIAA? Just because the police are allowed to batter down your door, doesn't mean you get sued for putting locks on it.
  • by Komrade S. (604620) on Sunday September 29, 2002 @02:53AM (#4352716) Homepage
    The /. effect!
  • by chris_7d0h (216090) on Sunday September 29, 2002 @03:04AM (#4352742) Journal
    Hmm. It seems this could be used to cut down on the time spent doing nitty-gritty stuff on each P2P project. "Gnutella like" projects might in the future spend more time doing GUI clients instead of implementing P2P specifications and protocols.

    It will be interesting to see which areas are going to be covered. I guess a way of identification will be included, so the "traditional" P2P projects will probably have to spend some time counter the identification stuff as well.
  • Or... (Score:4, Funny)

    by jpt.d (444929) <abfall@roge[ ]com ['rs.' in gap]> on Sunday September 29, 2002 @03:05AM (#4352743)
    "The Non-Slashdot-Fund has dumped a load of money to the IReallyIsn'tSlashdot project to research an anti-slashdot technology called Distributed Hash Tables as a tool for creating networks that are involnerable to slashdots. The chief purpose seems to be to stop Death-of-you-by-Slashdot attacks, from front page or otherwise."
  • Hash Tables (Score:2, Funny)

    by Anonymous Coward


    Well last time I was in Amsterdam they had these "Distributed Hash Tables". After about 2 hours I was feeling very decentralized.

  • Funny thing, last few years people have been calling client/server systems distributed systems. For me they are not distributed as the architect has simply decided to draw a line some where between the datafocused stuff (the provider) and the end user interface fluff (the consumer) and separated the logic into two components (often on separate nodes) The relationship on a conceptual level is always 1..* where many identical consumers use one and the same provider. For me distributed is P2P where a node can both provide and consume data centered services, rather than simply relying on others provide or consume (file-swapping p2p and the SETI apps are good examples).
    Anyway, that's my immediate reflection.
  • Just a thought... (Score:3, Informative)

    by mmoncur (229199) on Sunday September 29, 2002 @03:25AM (#4352783) Homepage
    a tool for creating networks that don't have "centralized points of vulnerability". The chief purpose seems to be to stop DoS attacks, intentional or otherwise.

    Wasn't that the goal of the ARPANET project that led to the Internet in the first place? I guess it didn't work.

  • Freenet? (Score:5, Interesting)

    by E1ven (50485) <e1venNO@SPAMe1ven.com> on Sunday September 29, 2002 @03:30AM (#4352786) Homepage
    Can anyone explain to me how this is different from Freenet? [freenetproject.org]

    Freenet has a Decentralized Architecture, which is specifically designed to resist DOS attacks, by making each client that views the page into a possible server..

    With freenet, any DDOS attempt would actually make the content MORE accessable, as it spread it to more and more nodes..

    • Re:Freenet? (Score:1, Flamebait)

      by mizhi (186984)
      The difference is that IRIS is developed by MIT and Freenet isn't.

      You can take that as either a sarcastic or serious comment. I think it's a little of both.
    • Re:Freenet? (Score:2, Informative)

      by T4D (602592)
      The primary goal of Freenet is anonymity. This means that there are some limitations/restrictions placed on the design. The IRIS project has some of the goals as Freenet, but they are not constrained by the additional goal of maintaining absolute anonymity. Without the need to maintain anonymity, they can focus on performance and scalability. They don't need to work about hiding the identity of the author and readers. Freenet is a great project and I hope that they can achieve all their goals. But, their project will never produce a general purpose distributed data storage solution. However, it looks like IRIS and related projects like Cord and CFS will do just that.
    • very. (Score:2, Informative)

      by jon_c (100593)
      god this got modded up? i swear, all someone has to do is mention freenet on slasdot, sigh..

      It's not like freenet. freenet searches work just like gnutella, randomly. it's a completely retarded way to organize a network. A distributed hash table like Circle [monash.edu.au] solves this by organizing the network in a logical, storable and efficient way.

      Basically compare a binary search vs. a random search, where the random search is like O(n) except you may just miss something.

      -Jon

      • Completely wrong (Score:4, Insightful)

        by Sanity (1431) on Sunday September 29, 2002 @04:36AM (#4352886) Homepage Journal
        You are sooooooo wrong, and it is you that have been modded up unjustly, since you clearly don't know what you are talking about.

        Freenet searches do not work like Gnutella, as you would know if you knew anything [freenetproject.org] about Freenet.

        Freenet's search has, through multiple independent simulation-based studies (cited in the link I give above), been demonstrated to have logarithmic scalability, not the linear scalability you claim.

        To inject some facts into this conversation - Freenet isn't exactly the same as a distributed hashtable, as it doesn't guarantee retrievability of information, but this is probably an inevitable consequence of achieving Freenet's goals, and Freenet's developers aren't shy about it.

        The claims you have made about Freenet are total FUD.

        • Freenet's search has, through multiple independent simulation-based studies (cited in the link I give above), been demonstrated to have logarithmic scalability, not the linear scalability you claim.

          The very paper you links to shows that median request path length is N^0.28. Logarithmic, that's not.

          Freenet has probabilistic, polynomial-time lookup and unbounded routing table size. Chord-like DHTs have deterministic, logarithmic-time lookup and logarithmic routing table size. Anonymity is nice, but it's costly.

          Freenet's worst-case performance -- i.e., when its routing table state is cold -- is O(N). Just like Gnutella. Chord's worst-case performance is still logarithmic.

          --Patrick

          • Achord [thalassocracy.org] is a DHT that can give some Anonymity protections.
          • The very paper you links to shows that median request path length is N^0.28. Logarithmic, that's not.
            The important thing is that it is sub-linear.
            Freenet has probabilistic, polynomial-time lookup and unbounded routing table size.
            Lookup of what? The routing table size is bounded.
            Freenet's worst-case performance -- i.e., when its routing table state is cold -- is O(N)
            You are using confused terminology. "Freenet" doesn't have a routing table, the individual participant nodes do.
            Chord's worst-case performance is still logarithmic.
            This worst case assumes that nodes in the Chord network don't fail, but given that nodes in any P2P network are prone to failure, this really isn't a worst-case at-all.
    • by Sanity (1431)
      While Freenet achieves many of these goals, there are at least two differences between Freenet and what most people would expect of a distributed hashtable:
      1. Freenet doesn't guarantee retrievability of information
      2. Freenet does its best to protect the anonymity of its users
  • by billstr78 (535271) on Sunday September 29, 2002 @04:10AM (#4352842) Homepage
    Becuase this story was posted [slashdot.org] on Wednesday already. Not only that, but it was the very last story on the topic [slashdot.org] of the "Internet". Either the mods have short memory loss, or they did not bother to look at the first posting which mentioned [infoworld.com] the IRIS grant as well.

    NOTE to Mods: DifferentTitle != DifferentStory
  • by m0rph3us0 (549631) on Sunday September 29, 2002 @04:21AM (#4352860)
    All networks have some centralized point of vulnerability. I really fail to see how distributed hash tables will reduce the vulnerability. The internet is probably the most decentralized network in the world, and even it has very vulnerable points. Someone attacking the internet would have a hard time taking it down, completely. However taking it down to a point of general unusability is within feasability given the current situtation and architechure. Today's internet has some very vulnerable points, namely the clients. A worm written in java that abused multiple platform security hole could easily render the internet unusable. I really fail to see how any sort of research is going to solve this problem. A better approach IMHO would be to
    1) elminate completely the use of plaintext protocols
    2)Start using our CPU's to the fullest using the computer to gaurd against software failure. One easy step would be to eliminate decrementing stacks. It's alot hard to overflow into a a memory space behind you then ahead of you.

    Just my 2 cents
    • Today's internet has some very vulnerable points, namely the clients.

      That is the upside of technologies like DRM. Hardening the clients with DRM and use of DHT would make the internet a lot more robust.

    • What about a different kind of abuse, like AOL's grip on their instant messaging database of users and the subsequent monopoly. This is because of the routing! All requests have to get routed to a central server. Things would be different if we had a decentralized user location system, which these new routing tables are bringing!
  • by gojomo (53369) on Sunday September 29, 2002 @04:33AM (#4352878) Homepage
    DHTs are also the key to the next generation of efficient, centerless P2P file-sharing.

    Two well-known academic DHT projects are Chord [mit.edu] and Kademlia [nyu.edu].

    Kademlia is the basis for VarVar [varvar.com] and EDonkey's successor, Overnet [overnet.com]. There's an experimental effort to add a Chord-style query routing option to Gnutella, to find exact files over the whole network with far less traffic.

    • It may also be good to mention that CHORD is being developed by Limewire [limewire.org] here [limewire.org]. A release is being expected any day now... should be pretty soon, as they've released their spec for GUESS [yahoo.com] in the GDF [yahoo.com], which is a global search method.

      With GUESS and CHORD, perhaps Gnutella can finally stand up to FastTrack, WinMX etc.
    • by Fzz (153115)
      The best known DHTs are: Now look who the NSF are funding for IRIS: MIT, ICSI, UC Berkeley, NYU, Rice. All the main DHT people in one project. That's got to be good.

      - Fzz

  • by RQuinn (521500)
    I'm certain this is just a clever move by those NSF terrorists to facilitate their upcoming anti-UNATCO plots!
  • A quick Google search reveals these tidbits on DHT vulnerabilities:

    Security Considerations for Peer-to-Peer Distributed Hash Tables [mit.edu]
    Achilles Heel of the DHT [cubicmetercrystal.com]
  • by nice (144965)
    You may stop prevent conventional DoS attacks, but denials of service will never die. Just give it a reason to adapt.
  • All the major DHT groups are involved. I wonder which DHT they're going to use.....

    Rice: Pastry (n-Hypercube) [nec.com]
    MIT: Chord (Ring-based) [nec.com]
    Berkeley: Tapestry [nec.com]
    ICSI: CAN (Mesh-based) [nec.com]
    • > I wonder which DHT they'll use

      The nice thing about DHTs is that the interface is nearly identical on all of the platforms: Given a key, find the associated object. (And insert, of course). Most of the DHT teams are already working together to create a common interface so that they can easily be evaluated against each other. It's likely that the higher-level results from IRIS will be DHT agnostic. Some of the lower-level things (like making the DHTs themselves more resilient) will probably be done using each group's own DHT.

      (Disclaimer: While I work in one of the groups that's participating in iris, these are only my guesses, not any kind of official word).
  • Roughly, this reminds me of David Gelernter's Linda tuple space (and also its progeny, JavaSpaces).

    Other posters have made interesting comparisons with Freenet, etc., but this research initiative seems (from the limited information in the linked project funding announcement) to be about persistent and replicated global data sharing. (Well, I guess that Freenet does that grin :-)

    For programmers: spend an evening or two playing with either JavaSpaces or IBM's distributed tuple space stuff - fun and educational if you havn't already checked it out.

    -Mark

  • As my Programming Languages instructor would say "Insert any drug references here." when discussing Hash Tables as something other than a table where you keep your hash.
  • This kind of system may be more resistant to technical problems like DoS, but I think that it's much more important for it to be resistant to political problems.

    No, I am not talking about legal problems like court orders. If the system as any central point there is the problem of who gets to control it. With no center it will be much easier for everyone to agree on the protocol without endless politicking.
  • so I see a lot of misconceptions about what the ITR is. People say it's like Freenet, or gnutella, no. DHTs are all "structured" decentralized peer to peer networks. That means there is a well defined routing algorithm between any node and any other node given a node ID in the system. What's more, it gives you the ability to deterministically find an object, unlike any of the existing p2p software applications.

    This is more about how to build large scale network applications such as multicast and file systems in an efficient way. DHTs like Tapestry (http://www.cs.berkeley.edu/~ravenben/tapestry), Pastry (http://research.microsoft.com/~antr/), Chord (http://www.pdos.lcs.mit.edu/chord/) and CAN are MUCH better at finding a single unique file than anything before. You can use it to find copies of britney spears, but that's not the point.

    Finally, these systems perform with contraints on performance, unlike Freenet. You're guaranteed to either find it or know it's not there (in the absence of failure corner cases) in a # of hops logarithmic to the size of the network. And all this is done without any super servers or supernodes to maintain the network. It is fully decentralized.
  • Last night the power went out. Good thing my camera had a flash....
    The neighbors thought it was lightning in my house, so they called the cops.
    -- Steven Wright

    - this post brought to you by the Automated Last Post Generator...

Nondeterminism means never having to say you are wrong.

Working...