Re: Packet loss when reaches 1Mb traffic

Hi again Elbert,

I forgot to mention, irrespective of what your NSP is doing, you
_must_ perform traffic shaping on the egress traffic. You have a
100Mbps interface (access) with a 2Mbps traffic contract.

An Ethernet interface knows how to transmit at only one speed -- in
your case, 100,000,000 bits per second. By shaping, you control for
how long (what fraction of) each second your interface will transmit.

If you don't have any special requirements, like latency sensitive
voice or video conferencing traffic, it could be as simple as:

policy-map SHAPER
 class class-default
  shape average 2000000
!
interface FastEthernet0/0
 description Link to NSP
 service-policy output SHAPER

From there, there are countless 'nerd knobs' to twist and it would be
impossible to cover all scenarios here.

I'm making some assumptions about the service your NSP is providing,
but if you are not shaping, it is almost certainly the cause of [the
majority of] your problems.

Cheers,
Dale

On Thu, Mar 11, 2010 at 7:02 PM, Dale Shaw <dale.shaw_at_gmail.com> wrote:
> Hi Elbert,
>
> On Thu, Mar 11, 2010 at 6:38 PM, Elbert Caibigan <xybert_gwapo_at_yahoo.com> wrote:
>>
>> I'm experiencing a problem (intermittent connection and choppy voice) with our
>> client. Our client is using a 2Mb leased-line via Metro Ethernet. Every time
>> the clients traffic reaches to 1MB they start experiencing intermittent
>> connection. Packet loss and abrupt increase of latency was observed. We check
>> the client interface and there is no error seen. Same on our interface (ISP
>> provider). No error seen on the port. We check the autonegotiation with the
>> leased-line provider and it was all set to FULL/100. The rate-limit configured
>> on the leased-line provider was 2MB for input and 2MB for output. We already
>> replaced the equipment of the client, cables and etc and change its port on
>> our side but still experiencing the same problem.
>
> I've seen lots of problems with carrier transmission equipment (e.g.
> SDH) dropping packets between your L3 device and theirs. In my
> experience, it's usually caused by relatively small buffers being
> quickly filled up at the point of interconnect.
>
> Does your NSP provide QoS on this service? If so, are markings
> observed/honoured in the L2 network or only at L3? Do you know what
> type of transmission network they have? How are they policing your
> traffic? (i.e. which exact parameters are they using for things like
> burst size or measurement interval and on what type of device?)
>
> In one extreme case, we were losing ~6Mbps of traffic between the CE
> and PE. Interfaces were running clean. Comparing the egress bitrate at
> the CE and the ingress bitrate at the PE told a different story, as
> did a packet sniffer.
>
> You might find if you send a steady stream of traffic at 2,000,000bps,
> it all makes it, but as soon as you throw in some bursty goodness, it
> all goes to mud.
>
> The moral of the story is you need to open a dialogue with your NSP
> and push them really hard to prove that they're not losing your
> traffic. Make sure your interface really is running clean (no drops,
> overruns, CRC/alignment errors etc.) before pointing the finger at
> them, though :-)
>
> cheers,
> Dale

Blogs and organic groups at http://www.ccie.net
Received on Thu Mar 11 2010 - 19:22:02 ART