Stories
Slash Boxes
Comments

SoylentNews is people

posted by janrinok on Friday July 18 2014, @02:46AM   Printer-friendly
from the faster-faster-faster dept.

MIT researchers are to present a new network-management system dubbed "Fastpass" that reduced the average queue length of routers in a Facebook data center by 99.6 percent virtually doing away with queues, according to experiments performed. Instead of using a decentralised method, Fastpass uses a central server called an "arbiter" to decide which nodes in the network may send data to which others during which periods of time.

With Fastpass, a node that wishes to transmit data first issues a request to the arbiter and receives a routing assignment in return. "If you have to pay these maybe 40 microseconds to go to the arbiter, can you really gain much from the whole scheme?" says Jonathan Perry, a graduate student in electrical engineering and computer science (EECS) and another of the paper's authors. "Surprisingly, you can." The researchers' experiments indicate that an arbiter with eight cores, or processing units, can keep up with a network transmitting 2.2 terabits of data per second. That's the equivalent of a 2,000-server data center with gigabit-per-second connections transmitting at full bore all the time.

"This paper is not intended to show that you can build this in the world's largest data centers today," Balakrishnan says. "But the question as to whether a more scalable centralized system can be built, we think the answer is yes." Moreover, "the fact that it's two terabits per second on an eight-core machine is remarkable," Balakrishnan says. "That could have been 200 gigabits per second without the cleverness of the engineering."

The key to Fastpass's efficiency is a technique for splitting up the task of assigning transmission times so that it can be performed in parallel on separate cores. The problem, Balakrishnan says, is one of matching source and destination servers for each time slot. "If you were asked to parallelize the problem of constructing these matchings," he says, "you would normally try to divide the source-destination pairs into different groups and put this group on one core, this group on another core, and come up with these iterative rounds. This system doesn't do any of that."

Instead, Fastpass assigns each core its own time slot, and the core with the first slot scrolls through the complete list of pending transmission requests. Each time it comes across a pair of servers, neither of which has received an assignment, it schedules them for its slot. All other requests involving either the source or the destination are simply passed on to the next core, which repeats the process with the next time slot. Each core thus receives a slightly attenuated version of the list the previous core analyzed.

Related Stories

Math can Make the Internet 5-10 Times Faster 28 comments

Mathematical equations can make Internet communication via computer, mobile phone or satellite many times faster and more secure than today. Results with software developed by researchers from Aalborg University in collaboration with the US universities the Massachusetts Institute of Technology (MIT) and California Institute of Technology (Caltech) are attracting attention in the international technology media.

A new study uses a four minute long mobile video as an example. The method used by the Danish and US researchers in the study resulted in the video being downloaded five times faster than state of the art technology. The video also streamed without interruptions. In comparison, the original video got stuck 13 times along the way.

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 3, Interesting) by clone141166 on Friday July 18 2014, @03:38AM

    by clone141166 (59) on Friday July 18 2014, @03:38AM (#70599)

    Hooray, we have re-invented the Token Ring/FDDI network topology?

    • (Score: 2, Interesting) by Anonymous Coward on Friday July 18 2014, @04:09AM

      by Anonymous Coward on Friday July 18 2014, @04:09AM (#70605)

      Looks more like a central computer controlling the time slots with a release/stop sort of scheduling window. Instead of a time slot that each card is assigned to.

      If you can keep your control network unsaturated this could work very nicely to remove collisions.

      The affect could be similar to token ring though.

      • (Score: 1, Funny) by Anonymous Coward on Friday July 18 2014, @10:38AM

        by Anonymous Coward on Friday July 18 2014, @10:38AM (#70726)

        The affect could be similar to token ring though.

        Feeling like going round in circles? :-)

    • (Score: 3, Interesting) by AnonTechie on Friday July 18 2014, @10:44AM

      by AnonTechie (2275) on Friday July 18 2014, @10:44AM (#70730) Journal

      I think this is applicable here:

      The Twelve Networking Truths: http://tools.ietf.org/html/rfc1925 [ietf.org]
      Especially:

      (6) It is easier to move a problem around (for example, by moving
                      the problem to a different part of the overall network
                      architecture) than it is to solve it.

                      (6a) (corollary). It is always possible to add another level of
                                indirection.

      --
      Albert Einstein - "Only two things are infinite, the universe and human stupidity, and I'm not sure about the former."
  • (Score: 2, Interesting) by Anonymous Coward on Friday July 18 2014, @03:40AM

    by Anonymous Coward on Friday July 18 2014, @03:40AM (#70601)

    Looks like token window but with a arbiter deciding 'yeah you can use that window' instead of waiting for the token to show up.

    So the arbiter always holds the token then decides which server will do the work. Even though there is no 'token' per se. Just windows. And a controller to pass the work around.

    So long as you dont overwhelm your arbiter network this would work.

    Yeah this would minimize the amount of retry. Which is killer on networks with lots of computers.

    • (Score: 0) by Anonymous Coward on Friday July 18 2014, @10:41AM

      by Anonymous Coward on Friday July 18 2014, @10:41AM (#70728)

      Well, I guess if you overwhelm your arbiter network, you can solve this with an arbiter for the arbiter network. ;-)

      • (Score: 0) by Anonymous Coward on Friday July 18 2014, @01:37PM

        by Anonymous Coward on Friday July 18 2014, @01:37PM (#70776)

        How about an n-arbiter depth network where a sub-arbiter is created for an arbiter network if it is saturated?