[173345] in North American Network Operators' Group
Re: Hurricane Electric packet loss
daemon@ATHENA.MIT.EDU (Tim Heckman)
Tue Jul 22 14:24:47 2014
X-Original-To: nanog@nanog.org
In-Reply-To: <CFF47082.EA7EA%wolfgang.nagele@ausregistry.com.au>
Date: Tue, 22 Jul 2014 11:24:37 -0700
From: Tim Heckman <t@heckman.io>
To: "Wolfgang Nagele (AusRegistry)" <wolfgang.nagele@ausregistry.com.au>
Cc: "nanog@nanog.org" <nanog@nanog.org>
Errors-To: nanog-bounces@nanog.org
Hey Wolfgang,
I believe I may be seeing similar behavior but it's hard for me to
confirm. My network configuration is one that mtr doesn't support, so
I can't get a report when we're having issues. I don't have my transit
provided directly from HE, but rather through a provider who colocates
out of one of their facilities. So I'm not sure I could even directly
reach out to the Hurricane Electric NOC to get help.
We've been seeing the odd connectivity issues between HE FMT2 (Linode)
and AWS US-WEST-1 and US-WEST-2. It's a mixed combination of loss and
increased latency, both which cause some hiccups in some of our
WAN-based clusters. There have been times where the issues we've seen
have been attributed to a DoS attack directed toward a Linode
customer, but there have been quite a few networking events that seem
to have no relation to a known attack.
Thanks for reaching out to NANOG with this issue, it may have shed
some light on some of the issues we are seeing.
Cheers!
-Tim
On Tue, Jul 22, 2014 at 2:48 AM, Wolfgang Nagele (AusRegistry)
<wolfgang.nagele@ausregistry.com.au> wrote:
> Hi,
>
> We=E2=80=99ve been customers of Hurricane Electric for a number of years =
now and always been happy with their service.
>
> In recent months packet loss on some of their major routes has become a v=
ery common (every few days) occurrence. Without knowledge of their network =
I am unsure what=E2=80=99s the cause of it but we=E2=80=99ve seen it on the=
Tokyo - US routes as well as the London - US routes. It reminds me of the =
Cogent expansion which was carried out by unsustainable oversubscription wh=
ich eventually resulted in unusable service for a number of years. Having s=
een some of the rates that HE has been selling for I can=E2=80=99t help but=
wonder if they made the same mistake ...
>
> Here is an example of what=E2=80=99s going on again atm.
> HOST: prolocation01.ring.nlnog.ne Loss% Snt Last Avg Best Wrst St=
Dev
> 1.|-- 2a00:d00:ff:136::253 0.0% 11 0.3 0.3 0.3 0.4 =
0.0
> 2.|-- 2a00:d00:1:12::1 0.0% 10 0.7 0.8 0.7 1.1 =
0.1
> 3.|-- hurricane-electric.nikhef 0.0% 10 0.7 3.1 0.7 8.3 =
2.9
> 4.|-- 100ge9-1.core1.lon2.he.ne 0.0% 10 9.8 12.6 8.0 19.2 =
4.1
> 5.|-- 100ge1-1.core1.nyc4.he.ne 10.0% 10 74.7 74.6 73.7 80.8 =
2.3
> 6.|-- 10ge10-3.core1.lax1.he.ne 30.0% 10 133.4 138.0 133.4 145.1 =
4.8
> 7.|-- 10ge1-3.core1.lax2.he.net 20.0% 10 135.7 139.1 133.4 145.1 =
4.5
> 8.|-- 2001:504:13::3b 40.0% 10 143.2 143.1 142.1 144.4 =
0.8
> 9.|-- 2402:7800:100:1::55 50.0% 10 144.4 144.1 143.8 144.4 =
0.2
> 10.|-- 2402:7800:0:1::f6 60.0% 10 298.7 298.4 298.2 298.7 =
0.2
> 11.|-- ge-0-1-4.cor02.syd03.nsw. 10.0% 10 299.3 298.9 298.3 299.5 =
0.5
> 12.|-- 2402:7800:0:2::18a 20.0% 10 299.7 299.4 298.9 300.1 =
0.4
> 13.|-- 2001:dcd:12::10 30.0% 10 299.8 299.5 298.8 300.0 =
0.5
>
> Is anybody else observing this as well?
>
> Cheers,
> Wolfgang
