Europe-to-US congestion and packet loss on he.net network, and their NOC@ won't even respond

Daniel Suchy danny at danysek.cz
Sun Dec 1 11:16:20 UTC 2013


On 1.12.2013 11:49, Randy Bush wrote:
>>> Using a 1/10th of a second interval is rather anti-social.
>>> I know we rate-limit ICMP traffic down, and such a
>>> short interval would be detected as attack traffic,
>>> and treated as such.
>> For what it is worth, I used to think the same, until I saw several 
>> providers themselves suggest that 1000 packets should be sent, with 
>> the 0.1 s interval.  So, this is considered normal and appropriate 
>> nowadays.
> 
> matthew is correct
> 
> go back to your old way of thinking.  while some providers may tolerate
> fast pings, few if any grown-ups do.  and even thouse who think they do
> have routing engines which consider all pings as low priority rubbish to
> be dropped when there is any real work to do.
> 

From router control-plane perspective, rate-limiting should be always
expected and result evaluation should take that in account. From router
perspective, packet with TTL=1 is handled typically in software, in CPU
with limited power (compared to modern hardware) and it's not a primary
job of router to answer to each TTL=1 packet - that's correct view.

But, provided reports shows ALSO end-to-end packet loss, which never
will be caused by control-plane policers on transit routers, these
packets will never hit router CPU.

And there we talk about basic network neutrality - everyone should treat
all data equally, independently of protocol used for data transport.

Daniel

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4240 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20131201/674166aa/attachment.bin>


More information about the NANOG mailing list