[Bloat] other people's VLAN congestion

[Bloat] other people's VLAN congestion

[Michael Richardson]

[-- Attachment #1: Type: text/plain, Size: 1268 bytes --]


This is for the page where we list places that you may have bufferbloat
and be unable to see it.  I think that there is a place in the wiki, but
I have to locate it, and I started writing this offline.

Many owners of (managed) fiber sell NNI to Access Port "LAN Extension"
services. Typically there is a 1GB/s ethernet NNI in the data center,
and a VLAN is assigned to travel to a 10Mb/s or 100Mb/s port at each site.

The consumer of this service is usually (Boutique) ISPs, but also larger
enterprises.   Without oversubscribing the NNI, the provider of the
fiber may have oversubscribed the service, and this can cause queues to
develop inside of this layer-2 network.  Even without oversubscription
in the provider network, instantaenous bursts can sometimes cause
significant jitter and buffer consumption.  

These buffers may have nothing to do with the consumer's
traffic. Addition of buffers here reduces the drop probability therefore
appears to the provider to be a good idea.

-- 
]               Never tell me the odds!                 | ipv6 mesh networks [ 
]   Michael Richardson, Sandelman Software Works        | network architect  [ 
]     mcr@sandelman.ca  http://www.sandelman.ca/        |   ruby on rails    [ 
	




[-- Attachment #2: Type: application/pgp-signature, Size: 307 bytes --]

Re: [Bloat] other people's VLAN congestion

[Ing-Jyh Tsang]

[-- Attachment #1: Type: text/plain, Size: 1499 bytes --]

Those queues can and should be correctly managed and configured. Both in 
terms of queue size and/or use of /RFC 2697/ 
<http://www.ietf.org/rfc/rfc2697.txt>
or /RFC 2698/ <https://tools.ietf.org/rfc/rfc2698.txt>, depend on each 
operator and what they are selling (SLA) ....

On 21/03/2013 15:53, Michael Richardson wrote:
> This is for the page where we list places that you may have bufferbloat
> and be unable to see it.  I think that there is a place in the wiki, but
> I have to locate it, and I started writing this offline.
>
> Many owners of (managed) fiber sell NNI to Access Port "LAN Extension"
> services. Typically there is a 1GB/s ethernet NNI in the data center,
> and a VLAN is assigned to travel to a 10Mb/s or 100Mb/s port at each site.
>
> The consumer of this service is usually (Boutique) ISPs, but also larger
> enterprises.   Without oversubscribing the NNI, the provider of the
> fiber may have oversubscribed the service, and this can cause queues to
> develop inside of this layer-2 network.  Even without oversubscription
> in the provider network, instantaenous bursts can sometimes cause
> significant jitter and buffer consumption.
>
> These buffers may have nothing to do with the consumer's
> traffic. Addition of buffers here reduces the drop probability therefore
> appears to the provider to be a good idea.
>
>
>
> _______________________________________________
> Bloat mailing list
> Bloat@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat


[-- Attachment #2: Type: text/html, Size: 2700 bytes --]

Re: [Bloat] other people's VLAN congestion

[Michael Richardson]

>>>>> "Ing-Jyh" == Ing-Jyh Tsang <ing-jyh.tsang@alcatel-lucent.com> writes:
    Ing-Jyh> Those queues can and should be correctly managed and
    Ing-Jyh> configured. Both in terms of queue size and/or use of /RFC
    Ing-Jyh> 2697/ <http://www.ietf.org/rfc/rfc2697.txt> or /RFC 2698/
    Ing-Jyh> <https://tools.ietf.org/rfc/rfc2698.txt>, depend on each
    Ing-Jyh> operator and what they are selling (SLA) ....

You'd think that, and the operator I know of is happy to sell an SLA for
more money.  But, the "best effort" service had the property that
packets were never dropped, but were regularly late, even though my edge
CPE equipment was prioriziting the way I wanted.

-- 
]               Never tell me the odds!                 | ipv6 mesh networks [ 
]   Michael Richardson, Sandelman Software Works        | network architect  [ 
]     mcr@sandelman.ca  http://www.sandelman.ca/        |   ruby on rails    [ 
	

Re: [Bloat] other people's VLAN congestion

[Ing-Jyh Tsang]

[-- Attachment #1: Type: text/plain, Size: 977 bytes --]


On 21/03/2013 17:53, Michael Richardson wrote:
> You'd think that, and the operator I know of is happy to sell an SLA 
> for more money. But, the "best effort" service had the property that 
> packets were never dropped, but were regularly late, even though my 
> edge CPE equipment was prioriziting the way I wanted. 

Rigth, in case of BE, not much is done, but mechanism are there. What 
are the incentives for the operators (you get what you pay for...)? 
However, as higher and higher access BW is offered/sold, this becomes 
more visible and operators I know does something about it, all the way 
to the CPE. I think the main question is can we improve beyond /RFC 
2697/ <http://www.ietf.org/rfc/rfc2697.txt> or /RFC 2698/ 
<https://tools.ietf.org/rfc/rfc2698.txt>? Specially with adaptive 
systems. On the other hand, many hesitated on the "adaptive", more 
"intelligence" systems siding/claiming control and stability... Due to 
lack of knowledge, mind frame, or ...

[-- Attachment #2: Type: text/html, Size: 1936 bytes --]