Little peak time latency humps

[#1]

Quote from: dave

if we can help minimise any impact of that traffic on their streaming or browsing then all the better.

Absolutely.
Quite dramatic, isn't it?

[#2]

So what's the next step then?
Doesn't seemed to have fixed the issue and I don't think a flat ping graph is too much to ask.

[#3]

Eh?  it has pretty much fixed the issue?  (we're just experimenting before committing to the change?)

[#4]

Took me literally all day to get on PCL-AG02 after BT's maintenance kicked me off yest  

The general ping spikes are an Asus bug....seems to just affect their routers, no idea why. Tried every setting but to no avail.

[#5]

The explanation from dave is very interesting, but it's got me thinking.
What exactly are the priorities when the Pro package is added?

[#6]

Pro puts everything into Gold except for gaming, voip, vpn into Titanium.

[#7]

I got me thinking too, as I'm not entirely convinced about putting ICMP into 'titanium', and I wonder whether the TBB graphs would reveal more interesting and useful results, if ICMP priority was 'bronze'.
I can see the point of view that says 'titanium' pings show what the line is capable of, will give relatively clean graphs, and artificially produce good looking statistical results.
However 'bronze' pings, would show 'you should get better than this' graph, which will be much more 'noisy', and the minimum latency will drift up and down depending on time of day and line loading.

[#8]

I should be able to set it up and see what it does on a test config. Not sure I'd want to put it anywhere other than titanium on the normal profiles because some things use ICMP to determine the latency and high pings could cause problems. I'll have a look next week after we've done testing next week.

[#9]

Dave,
Does the traffic shaping or something on your network add any significant latency?
I only ask as I have neighbours on different FTTC ISPs (One TalkTalk and one AA) and they both get around 7ms to say bbc.co.uk, whereas I seem to get around 11ms.

[#10]

[quote=dave]some things use ICMP to determine the latency and high pings could cause problems
Surely those applications (such as games) that are clever enough to require a latency calculation to provide the best interactive experience, then giving them a slightly worse 'bronze' ping delay would mean that because the actual application traffic is most likely to be of higher priority (e.g. 'gold' or 'titanium'), then the actual user experience should actually be better as the calculated latency will be based on worst case, rather than fastest achievable latency for the users connection.

[#11]

Quote from: Kelly

Eh?  it has pretty much fixed the issue?  (we're just experimenting before committing to the change?)

But pings are still fluctuating at peak time?
Surely you either have fluctuating pings or traffic management. Not both...
Take my Bethere line for the last 6 years with no traffic management:

If there was no time axis you'd never know when peak time was.

[#12]

Should be less than 1ms across our network by the looks of it, this is the ping to the BBC from PCL-AG01:
PCL-AG01#ping bbc.co.uk
Resolving "bbc.co.uk" ...
Sending 5 ICMP echoes to 212.58.251.195, timeout = 2 sec.
!!!!!    
Success rate = 100% (5/5), round-trip min/avg/max = 1/1/1 ms
PCL-AG01#
I think it rounds to the nearest ms.

[#13]

[What priority are ICMPv6 packets on? I see a marked difference in IPv4 compared to IPv6 ping graphs.] Scratch that, all IPv6 tunnel traffic goes through as IPv4 packets when concerning PN. So, modifying my question, what is the classification with IPv6-in-4 packets?

[#14]

Quote from: Amos91

Quote from: Kelly Eh?  it has pretty much fixed the issue?  (we're just experimenting before committing to the change?) But pings are still fluctuating at peak time? Surely you either have fluctuating pings or traffic management. Not both... Take my Bethere line for the last 6 years with no traffic management: If there was no time axis you'd never know when peak time was.

The change was only on 2 gateways, and was backed out this morning.

[#15]

@'dave' -
So what was the difference between this weeks fixes on PCL-AG01 and PCL-AG02 ?
I ask as my connection was never heavily loaded during peak time this week, and the PCL-AG01 'fix' didn't do anything to my poor TBB graph on Monday, but the PCL-AG02 'fix' flattened the blue average latency 'hump' on my plots on Tuesday and Wednesday.
Interestingly now that you have removed the 'fix' on PCL-AG02, my peak time 'hump' has returned, today Thursday.