[ndnSIM] leaky bucket start time
alexander.afanasyev at ucla.edu
Wed Feb 13 11:30:07 PST 2013
I see. Yeah, 1 second randomization could be a bad choice (any small randomization would be sufficient), but I wouldn't rely too much on users to take care of that.
I can actually put a configurable option to enable/disable this randomization, as well as to configure randomization interval, keeping it ON by default with 1 ms randomization interval.
On Feb 13, 2013, at 11:23 AM, Yaogong Wang <ywang15 at ncsu.edu> wrote:
> I see your point. Such kind of synchronization could lead to some simulation artifacts. But I believe this should be taken care of by the users. They should be aware of this (such synchronization problems could happen in any simulator) and incur some randomness in their simulation setup.
> My concern with the current implementation is that the leak start time is randomly picked between 0s and 1s. It is fairly possible that lots of packets have already been sent before the leak start time. These packets may get stuck in the network because the bucket didn't start leaking when it should have. This might not be a big issue most of the time but did cause me some headache in some particular setup.
> On Wed, Feb 13, 2013 at 1:57 PM, Alex Afanasyev <alexander.afanasyev at ucla.edu> wrote:
> True. But if two guys receive a packet at the same time (can happen in some simple topologies), they would start leaking at the same time, resulting in synchronized decision, which may lead to unexpected results (I had them with simplistic topologies).
> On Feb 13, 2013, at 10:39 AM, Yaogong Wang <ywang15 at ncsu.edu> wrote:
>> It doesn't seem to make much sense. The bucket anyway starts out as empty. Leaking won't really happen until someone calls BorrowLimit().
>> On Wed, Feb 13, 2013 at 1:17 PM, Alex Afanasyev <alexander.afanasyev at ucla.edu> wrote:
>> I've put it there to ensure that nodes don't make a synchronized leaks of buckets, which impossible in real life, but could happen in some cases in the simulations.
>> On Feb 13, 2013, at 9:45 AM, Yaogong Wang <ywang15 at ncsu.edu> wrote:
>> > Hi,
>> > I'm curious why do we randomize the leak start time in LimitsRate::NotifyNewAggregate() under untils/ndn-limits-rate.cc. Why can't we just start leaking at 0.0?
>> > Thanks,
>> > Yaogong
>> Yaogong Wang, PhD Candidate
>> Department of Computer Science
>> North Carolina State University
>> ndnSIM mailing list
>> ndnSIM at lists.cs.ucla.edu
> Yaogong Wang, PhD Candidate
> Department of Computer Science
> North Carolina State University
> ndnSIM mailing list
> ndnSIM at lists.cs.ucla.edu
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the ndnSIM