perfSONAR User Q&A and Other Discussion

["Kinkaid, Kyle" <>]

Hi All,

For what it's worth, I'm working on a project to impliment perfSonar on Odroid C2s.  Odroid C2s are SBCs, similar to RPi3s (they use the same ARM processor) and have a GigE port (which can push a full gig).  We are having issues with perfSonar but it remains to be seen how much of that is caused by the hardware, how much is caused by our choice of OS (Ubuntu...not officially supported by perfSonar), and how much is caused by us just not getting the setup and config right.

I look forward to putting together a lessons learned after I get it working (or give up) for the community benefit.

So far, I've been impressed by the C2 hardware and although it's lacking in several places, the network performance on them has been quite good.

-Kyle

On Tue, Feb 28, 2017 at 2:13 PM, Chevalier, Scott S <> wrote:

Just throwing more comments for consideration. The pi-class devices are
generally not recommended by the perfSONAR community for a number of
reasons documented elsewhere and by others already having responded. There
are users of the platform incorporating pi-class gear for
experimental/playground purposes and for very specific tasks where the
limitations are understood and accepted. YMMV

Alan Whinery (University of Hawaii) has assembled a great deal of data on
using pi-class devices and his postings and some data can be found at
http://perfclub.org/

In fact, this link - http://perfclub.org/?p=190 - goes directly to a post
from today recounting some of Alan¹s work on ³the SWARM² (his pi-class
campus-wide performance monitoring exeriment) including the pros, cons,
and deviations from the norm which can be expected when working on
pi-class hardware.


Scott Chevalier
Network Systems Analyst
International Networks @ IU <>
812-856-9964






On 2/28/17, 15:43, " on behalf of
Matthew Woodward" < on behalf of

> wrote:

>Also, you'd be limited to a max of 480Mbps (although it would probably
>be less than that, say, ~300Mbps) since USB2 tops out at that speed.
>
>- M
>
>On 2/28/17 3:16 PM, Steven Wallace wrote:
>> I¹m planning something similar. One issue that others have raised is
>>that the pi¹s ethernet interface uses a USB controller, resulting in a
>>fair amount of jitter for delay testing.
>>
>>
>>
>>
>>
>>> On Nov 21, 2016, at 4:13 PM, Ryan Harden <> wrote:
>>>
>>> Hello,
>>>
>>> I¹m digging up an old topic here, (yes I¹ve read the archives) but I¹m
>>>curious what the current state of affairs is. The most recent
>>>conversations I can find date back to 2014.
>>>
>>> I¹m working on building a low-cost perfSonar mesh to measure loss and
>>>latency (throughput isn¹t a requirement) across our campus. My goal is
>>>to build something that can be deployed quickly and easily in either a
>>>permanent or temporary fashion. Predictably, the RPi3 looks very
>>>attractive for this, so I¹ve acquired four to experiment with. I¹m not
>>>looking for ns latency resolution, just obvious spikes in latency that
>>>something like an RPi can detect in a reliable way.
>>>
>>> My goals are:
>>> 1) Power via PoE (done with adapter)
>>> 2) Run basic perfSonar packages (perhaps rebuilt for ARM?)
>>> 3) Gather wireless connection/auth statistics.
>>> 4) Run with zero local disk (PXE boot? NFS?) (Cycle PoE port triggers
>>>software upgrade, etc)
>>> 5) Run periodic latency/loss tests. (think 30 second 1Mbps UDP tests,
>>>nothing crazy)
>>>
>>> Has anyone already worked on this? Any notes specifically for the
>>>RPi3? It¹s significantly faster and has some upgraded hardware over the
>>>original/RPi2 versions.
>>> Given the relatively low investment so far, I could reset and go for a
>>>small x86 device, but I¹d prefer to stick with the RP if I can.
>>>
>>> Thanks
>>>
>>> /Ryan
>>>
>>> Ryan Harden
>>> Research and Advanced Networking Architect
>>> University of Chicago - ASN160
>>> P: 773.834.5441
>>>
>