[ndnSIM] Network Throughput

[Alex Afanasyev]

On Dec 20, 2013, at 5:58 AM, Mohammad <Mohammad.Hovaidi.Ardestani at aalto.fi> wrote:

> Hi Alex,
>  
> In my dumbbell scenario, there are three Interest senders that are connected to the first router (Rtr1). The bottleneck is a 1Mbps link which is based on connection of Face 3 of Rtr1 and Face0 of Rtr2.
> I run a ndn-simple scenario, where I have added L3RateTracer (ndn::L3RateTracer::InstallAll ("L3-rate-trace.txt", Seconds (0.5));) and modified frequency of interests to be 100 for every data requesters.
>  
> Time    Node    FaceId  FaceDescr   Type    Packets Kilobytes   PacketRaw   KilobytesRaw
> ...
> 4.5   Rtr1  3     dev[3]=net(3,3-4) InData      118   121.798     59    60.9004
> 5     Rtr1  3     dev[3]=net(3,3-4) InData      118   121.799     59    60.8994
>  
>  
> I have some questions:
>  
> 1.       Why are you considering only InData?  (If for instance we consider OutData of Rtr2 the result would be more unclear.)

Hi Mohammad,

InData trace guarantees that this amount of Data was actually transferred over the channel. OutData trace tells you the amount of Data NDN layer send out towards the channel, but there are no guarantees that this data was actually transmitted.  That is, if you're trying to send more than the available bandwidth, then some of the transmitted (and counted towards OutData) would be dropped.  If you're using point-to-point links, the dropped data is tracked under DropData trace.

> 2.       I reckon the throughput is KB per second, therefore the actual throughput of the bottleneck (in case of just considering InData of one router) should be twice the result shown above. 

Why twice?  The value in Kilobytes column is adjusted to the averaging period, so it always shows the estimated throughput in KB per second.  KilobytesRaw on the other hand, shows aggregated amount of kilobytes that was send out within the period.

---
Alex

> I really appreciate your help in advance!
>  
> BR
> -Mohammad
>  
> From: Alexander Afanasyev [mailto:cawka1 at gmail.com] On Behalf Of Alex Afanasyev
> Sent: Wednesday, December 18, 2013 11:42 PM
> To: Mohammad
> Cc: ndnsim
> Subject: Re: [ndnSIM] Network Throughput
>  
> Hi Mohammad,
>  
> I run a ndn-simple scenario, where I have added L3RateTracer (ndn::L3RateTracer::InstallAll ("rate-trace.txt", Seconds (1.0));) and modified frequency of interests to be 1000.
>  
> The resulting rate-trace.txt is consistent with the results in the visualizer and shows almost full utilization of the 1Mbps channel.  Here is a selected results from the trace (only for Node 0/face0 and InData parameter):
>  
> Time    Node    FaceId  FaceDescr   Type    Packets Kilobytes   PacketRaw   KilobytesRaw
> ...
> 17  0   0   dev[0]=net(0,0-1)   InData  117.834 122.087 118 122.259
> ...
>  
> As you can see, it shows that the estimated rate is 122.087 Kilobytes/s, which is 976.696 Kilobits/s which is 1000136 bytes/s (this value is not 100k, because there is estimation and  errors).
>  
> ---
> Alex
>  
> On Dec 13, 2013, at 3:20 AM, Mohammad <Mohammad.Hovaidi.Ardestani at aalto.fi> wrote:
> 
> 
> Hi Alex,
> I still have problems with finding link throughput. In 1Mbps link I get the 1300Kbytes throughput based on your explanation. I guess the link throughput should not be greater than 250Kbytes. Is there any other metric to get the channel throughput?
>  
> Thanks!
>  
> From: Alexander Afanasyev [mailto:cawka1 at gmail.com] On Behalf Of Alex Afanasyev
> Sent: Wednesday, August 21, 2013 9:05 PM
> To: Hovaidi Ardestani Mohammad
> Subject: Re: [ndnSIM] Network Throughput
>  
> Yes. It should be a rough equivalent (at least I'm using it in my scenarios), which excludes any layer-2 overhead.  Just make sure you summing up based on individual faces, and distringuish between In and Out packet types.
>  
> ---
> Alex
>  
>  
> On Aug 21, 2013, at 11:01 AM, Hovaidi Ardestani Mohammad <mohammad.hovaidi.ardestani at aalto.fi> wrote:
> 
> 
> 
> Hi Alex!
> What if we add throughputs off all packet types together? Is the result equal to throughput of channel or at least reasonable approximation of it?
> I need to have some parameters to measure the performance of the technique used to control congestion.
> Thanks again!
>  
> From: Alexander Afanasyev [mailto:cawka1 at gmail.com] On Behalf Of Alex Afanasyev
> Sent: Wednesday, August 21, 2013 7:25 PM
> To: Hovaidi Ardestani Mohammad
> Cc: ndnsim at lists.cs.ucla.edu
> Subject: Re: [ndnSIM] Network Throughput
>  
> Hi Mohammad,
>  
> ndn::L3RateTracer provides several metrics, including throughput estimation for each packet type (the 7th column "Kilobytes").  It could be not exactly the throughput in the channel (for that you would need to extend L2RateTracer implementation, which is now very limited), but in most cases is a very good approximation of the throughput with breakdown by NDN packet types.
>  
> ---
> Alex
>  
> On Aug 21, 2013, at 8:59 AM, Hovaidi Ardestani Mohammad <mohammad.hovaidi.ardestani at aalto.fi> wrote:
> 
> 
> 
> 
> Hello,
> I am wondering how it is possible to measure network throughput in ndnSIM scenario. The number of satisfied interest packets are given in l3-rate-tracer , but I am not quite sure whether it is the actual throughput or at least it can help to find it.
> Thanks for your response in advance.
> -Mohammad

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.lists.cs.ucla.edu/pipermail/ndnsim/attachments/20131222/40082c91/attachment.html>