Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

What sort of speeds are you getting in the evening peak then Tony? My 50Mb is currently only getting anything between 5-12Mb/s at peak times since a few weeks, but the graph is nowhere near that bad.

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

From what I've experienced then TBB is skewed more towards showing upstream utilisation and mine doesn't indicate downstream utilisation to anything like the same extent.
(Since I have Samknows then I have a relatively accurate and consistent benchmark to compare against)

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

That's less about the TBB graphs than it is about latency on DOCSIS simply being affected more by upstream utilisation than down.

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Two issues at play here.

1) DOCSIS upstream is TDMA which is more prone to latency caused by utilisation - there are 2 contention points at play, contending to send your request for a grant and then waiting for the grant time to actually transmit the data.
2) DOCSIS allows large traffic bursts to request more bandwidth as part of a data burst so favours continuous large amounts of upstream traffic over small bursts such as a response to a TBB ping.

Downstream uses a scheduler at the CMTS. The pipe has to max out, with interest, in order to increase latency substantially.

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Which makes sense but what about the relationship between download speed and latency.

At peak times my download drops to 50% (50Mb/s from 100Ms/s) but the peak latency on the graph does not increase significantly (outside of the test window).

If my download packets cannot be scheduled at the required rate to achieve 100Mb/s and are buffered in the CMTS then surely ICMP packet would be delayed too?

Also, the TBB speed test during peak hours shows a big drop in the x1 speed even before the x6 rate gets reduced. This x1 rate always starts high (close to 100Mb/s) and drops back to 20M/s or so.
To my mind this suggest some packet priority scheme on the downstream based on how close together packets are received by the CMTS for a particular CPE.
So if you're gaming with infrequent bursts these get scheduled immediately but if you are downloading the packets get delayed.

This is my speed test at ~11am

https://www.cableforum.co.uk/images/...2015/02/15.png

Ian

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

ICMP does not require 100Mb/s. There may not be enough capacity left to schedule an additional 100Mb/s while still having 0.01Mb/s required for ICMP.

---------- Post added at 12:33 ---------- Previous post was at 12:28 ----------

Oh how I wish VM were actually that competent.

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Easy - the DOCSIS scheduler uses something similar to Weighted Fair Queuing.

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Had to look that up but from Wiki...

'...an ill-behaved flow (who has sent larger packets or more packets per second than the others since it became active) will only punish itself and not other sessions' which is kind of what I described (in a less technical way).

Do the Virgin CMTS really implement that? On the upstream too?

Ian

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Downstream it works that way, upstream doesn't use the DOCSIS scheduler it uses TDMA.

See my earlier post - http://www.cableforum.co.uk/board/35...-post4910.html

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

I guess something similar could be implemented on upstream by limiting the rate of the 'grants' for 'ill-behaved' streams.

Thinking about the downstream, this definitely seems to explains why the TBB graph can look quite good even though download speeds have dropped significantly due to congestion; infrequent ICMPs being scheduled in advance of download packets for other CPEs.

I would have thought this constitutes a form of traffic management / priority scheme?

Ian

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

The scheduler isn't a traffic management scheme, it doesn't really have any intelligence to speak of.

I was, perhaps, overly kind when I described the scheduler as being WFQ. It doesn't have the smarts to 'selectively' slow heavy bandwidth flows, it's more that it shares bandwidth between flows and the way it empties buffers.

Essentially smaller packets, such as pings, get through more quickly than larger ones, such as a standard download.

On the upstream you can, certainly, prefer certain traffic, indeed it's needed as part of voice services, however it's of questionable value and is more inefficient in terms of usage of the resource to prefer pings or whatever over normal traffic.

This will be useful in DOCSIS 3.1 though and will I suspect present way more of a benefit than playing with upstream schedulers.

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

https://www.cableforum.co.uk/images/...2015/02/29.jpg

Well that's a very up-to-date game...

---------- Post added at 13:59 ---------- Previous post was at 13:55 ----------

[Edit]

So basically they've implemented random early detection, a standard, widespread queue management mechanism that's been around since 1993...

... And what about ECN? Why aren't more people using it?

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

I seem to be getting awful jitter at the moment!

https://www.cableforum.co.uk/images/...2015/02/14.png

Re: Think Broadband Ping Monitor Results (POST YOURS)
 

Well as you can see, there's absolutely nothing wrong with my area at the moment.

http://www.thinkbroadband.com/ping/s...03-02-2015.png