[ixpmanager] SFLOW Under Reporting?

[Nicolaas van der Ven (Nokia)]

Hi Ian,
We where working on extending and optimizing the flow collection and created a PR for it, so far we are blocked by understanding how to run this additional collector/parser.
See previous communication below.

So we only need some help from the community on this one to get started and take it up from there and I hope this will also help you and the rest of the community.
We have the resources available to extend and optimize this great project.


#########################################################
We have a proposal to enhance IXP manager by enabling additional forms of traffic flow information, illustrated below

[cid:image001.png at 01D992E1.88FD2A50]

'goflow2' (https://github.com/netsampler/goflow2) is an open source flow collector tool that supports multiple input formats, and produces JSON structured output.
At https://github.com/inex/IXP-Manager/pull/836/files you can find a draft PR outlining what this could look like.

I have a question about running sflowtool in the context of the Docker-compose (https://docs.ixpmanager.org/dev/docker/), see https://github.com/inex/IXP-Manager/issues/837 (closed issue).
Would appreciate any suggestions/pointers on how to test sFlow support and the new Perl script mentioned above

Also, we would like to hear from existing IXP manager users if this extension would benefit them - do you have devices that are currently not integrated due to the lack of sFlow support?

Thanks,
Jeroen van Bemmel, Nokia
#########################################################

Niek van der Ven
Nokia Webscale EMEA
Mobile: +31639634590   
Email: nicolaas.van_der_ven at nokia.com

Ian Chilton wrote on 24/06/2023 09:51:
> Hi,
>
> On 2023-06-24 09:50, Nick Hilliard (INEX) wrote:
>> are you running rrdcached?
>
> Yep!

for reference, inex's cache flushes take ~2-3 seconds. It may be happening that you're i/o bound on your RRD spool disk. If you have cache flushes which take longer to run than the flush timeout, for sure this would cause problems.? What's the i/o load looking like on the disk? Assuming your partition is /dev/sda$something, e.g.

# iostat --dec=0 -y sda 1 | grep --line-buffered sda

The first column will give TPS with 1s granularity.

Reloading the mac table should be pretty quick. This is just a structured dump of an SQL query.

Nick