Re: Google Cloud and IX - Traffic behavior

[Alain Hebert <ahebert () pubnix net>]

    Thanks Tom,

    Yeah my test points VM's where all FreeBSD 10.3/11.0, so I decided 
to randomize some of them and added a CentOS on my Telia peer (10Gbps), 
and start getting normal performance from GCLD ( In the 500Mbps/500Mbps 
range )

    PS: Even weirderer(tm) The *BSD VMs always performed correctly in 
the past, but not for this particular case.

    I have some work left to normalizing those peers, the same tactic 
didn't work on the other ones.

    if anyone have some lead on a TCP stream analyzer that would cut 
down on the number of naps I need to take looking for clues in wireshark 
captures.

-----
Alain Hebert                                ahebert () pubnix net
PubNIX Inc.
50 boul. St-Charles
P.O. Box 26770     Beaconsfield, Quebec     H9W 6G7
Tel: 514-990-5911  http://www.pubnix.net    Fax: 514-990-9443

On 06/21/17 11:12, Tom Beecher wrote:
> Just did a quick test from a personal VM, no throughput difference 
> over direct peering, public IX, or transit. GCP might have a 
> bottleneck in your case though, might be a good idea to ask them.
>
> Also, I'll have what Gordon is having.
>
> On Tue, Jun 20, 2017 at 8:49 PM, Gordon Cook <cook () cookreport com 
> <mailto:cook () cookreport com>> wrote:
>
>
>
>     Hi Alain and all the rest
>       I het it now
>
>     no offense and alain no harm done and all of nanog thank you and i
>     will continue observing as i have since 1995
>
>     thank you allagin
>     PS and BTW i am interested in CLOUD
>
>     :-)
>     thanks once more
>
>
>
>
>     > On Jun 19, 2017, at 9:36 AM, Alain Hebert <ahebert () pubnix net
>     <mailto:ahebert () pubnix net>> wrote:
>     >
>     >    Hi,
>     >
>     >    Yes Stephen, we're talking the usual like GTT...
>     >
>     >    And no latency wise they're about the same. In the 35ms range.
>     >
>     >    But I still can't figure out the 10 x drop, that level of
>     latency alone cannot be the factor.
>     >
>     >    ( And Gordy...  what?!? )
>     >
>     > -----
>     > Alain Hebert ahebert () pubnix net <mailto:ahebert () pubnix net>
>     > PubNIX Inc.
>     > 50 boul. St-Charles
>     > P.O. Box 26770     Beaconsfield, Quebec     H9W 6G7
>     > Tel: 514-990-5911 <tel:514-990-5911> http://www.pubnix.net  
>     Fax: 514-990-9443 <tel:514-990-9443>
>     >
>     > On 06/16/17 19:42, Stephen Fulton wrote:
>     >> Alain,
>     >>
>     >> When you refer to "normal peering" do you mean Internet
>     transit? Or are these PNI's with Google?  Do the GCLD instance you
>     reach through "normal peering" have higher latency than through TorIX?
>     >>
>     >> -- Stephen
>     >>
>     >> On 2017-06-16 6:58 PM, Alain Hebert wrote:
>     >>>     Hi,
>     >>>
>     >>>     Anyone aware of different traffic behavior depending if
>     the target goes through normal peering than through an exchanges
>     google exists in?
>     >>>
>     >>>     We're facing a weird issue where the same GCLD Instance
>     can upload up to 200Mbps (Ref 1) if the target path goes through,
>     lets say TorIX, but cannot get more than 20Mbps on similar hosts
>     (8 of them) sittings on our peering links.
>     >>>
>     >>>     PS; Those sames hosts get up to their link limit ( 1Gbps )
>     between each others and others test points we have;
>     >>>
>     >>>     PS: Wireshark capture show nothing abnormal;
>     >>>
>     >>>     PS: Links aren't congested, and so on...
>     >>>
>     >>> Ref 1 - 200Mbps is on a link rate-limited to 300Mbps.  Its my
>     only test point with a TorIX access
>     >>>
>     >>
>     >
>     >