Thanks James, Useful information, I have customers with some Sky circuits so will do some testing on those as well to compare.
Best regards, Paul From: James Bensley <[email protected]> Sent: 22 July 2022 08:20 To: [email protected]; Paul Bone <[email protected]> Subject: Re: [uknof] Policing vs Shaping in the UK Hi Paul, ------- Original Message ------- On Monday, July 11th, 2022 at 12:46, Paul Bone <[email protected]<mailto:[email protected]>> wrote: One of my ISP customers has multiple 2Gbps national ethernet circuits through two different providers. Testing the circuits of Provider A gives UDP in excess of 1.8Gbps with no packet loss, and TCP up to 1.8Gbps in both directions. This sounds like everything is in order. UDP testing really is the way to go here IMO because TCP is so varied, it's completely unpredictable and unreliable in terms of repeatable testing between operating systems and between applications. However, with provider B, I am seeing UDP packet loss when going over 1.5Gbps and TCP of less than 1Gbps even as low as 600M for upload tests (Download tests same as provider A). This needs to be tested with UDP and at multiple times throughout the day to account for any congestion the carrier may be having. It's hard to highlight just how useless TCP testing is IMO (although I understand, sometimes all you've got to hand it speedtest.com or similar). You say you see packet loss with UDP from 1.5Gbps onward, in which case, this is a problem with the provider, no question about it. If you are paying for 2G, you should be getting it with UDP (less overheads of course). This is consistent across multiple circuits and leads me to believe that Provider B is using a download shaper but an upload policer which is causing significant issues with circuit speeds. I read you results the other way around; sounds like they have egress shaping at the NNI end which gives the poor upload results, but egress policer at the remote site end, which is why download is fine with UDP. I have always tried to shape on egress at either end of a circuit wherever possible as ingress policers can be somewhat aggressive, but just wondering how common the use of policers is within the UK network operators? As Mark has already said, I think most telco's prefer policers otherwise you're adding latency into your customers traffic. Most national Ethernet services are providing a "dump" L2 pipe. Any decent provider should set the policer CIR to the rate you're paying for so you should be able to achieve it easily with UDP testing. If you want to then add shapers and priority queuing etc to your devices at each end, you can, but the provider should provide a dump pipe with guaranteed bandwidth. Also I think ingress policers at each end of the service is more common. This is the way it is with Sky Ethernet for example. Egress policers or shapers means that excess traffic is carried all the way across the network, just to be dropped, so you really want ingress to stop wasting bandwidth. Also most devices only support policers on ingress (you can't control what someone else sends you, so implementing ingress shapers is hard), and policers or shapers at egress, but if you're limiting at ingress at both ends you don't need any egress limiting. Cheers, James.
