BQM - I think not relevant to gaming performance on PN

Like many others I have the evening BQM hump even when not using the computer. I get the impression some people take this to indicate the likelihood of poor gaming performance.

I don't think this is the case. My point being that the PN traffic management is specifically designed to prioritise gaming traffic, and bog standard ping tests are almost certain to be in the bronze queue.

Agreed, theoretically, it should work in Plusnets favor.

My personal experience of Plusnet is that, from time to time, I experience appalling latency (120ms+) which has resulted in dire VOIP call quality and impossible gaming. I logged a call with Plusnet ages ago and they basically said they can't do anything.

I experimented with reconnecting to another aggregator and so far this has worked and restored really excellent latency.

I've experienced this type of behavior on both my DSL connections which are on different exchanges.

As an aside I noticed that since they did some work on the aggregators my latency increased by about 3-5ms but it differs from one aggregator to another.

Good point.

You should be able to see the queue used if you get some Wireshark logs of the packets. Not sure of the details though - but they can be found on the PN site.

In reply to a post by RobertoS:

I don't think this is the case. My point being that the PN traffic management is specifically designed to prioritise gaming traffic, and bog standard ping tests are almost certain to be in the bronze queue.

I got the impression from earlier discussions that bronze will only run slower than titanium on an idle line when Plusnet's pipes are full.

Oh lord - this is the last thing I want to read after just signing up to a 18 months this week!

I game online everyday, I'm doomed. I'm dooooomed!

In reply to a post by narz:

Oh lord - this is the last thing I want to read after just signing up to a 18 months this week!

I game online everyday, I'm doomed. I'm dooooomed!

lol. I'll play a few times a week (on xbox) and haven't noticed any issues... in fact...the missus will back me up with the fact that I can game and she can look at her next and debenhams websites without me bitching about the lag.

no value to this post really - just thought id try and cheer you up

In reply to a post by Elkin:

lol. I'll play a few times a week (on xbox) and haven't noticed any issues... in fact...the missus will back me up with the fact that I can game and she can look at her next and debenhams websites without me bitching about the lag.

no value to this post really - just thought id try and cheer you up

Good reviews from gamers definitely help, and are a plus point. he he.

When using games, I've only ever noticed it twice in the last 5 months - and it went back to normal after a quick reconnect to a different gateway - so even the fix is easy.

You probably want to see this page of TBB graphs:
http://community.plus.net/ping-graphs/

In reply to a post by narz:

Oh lord - this is the last thing I want to read after just signing up to a 18 months this week!

I game online everyday, I'm doomed. I'm dooooomed!

That was my point . That you are exactly the opposite from doomed. Note on the graphs page that WWWombat has just given you the link to that there is no packet loss.

In reply to a post by WWWombat:

You probably want to see this page of TBB graphs:
http://community.plus.net/ping-graphs/

I'm trying to find some recent O2 ping graphs to see how it compares.

Would you stress more importance on the maximum latency or the average latency?

In reply to a post by RobertoS:

That was my point . That you are exactly the opposite from doomed. Note on the graphs page that WWWombat has just given you the link to that there is no packet loss.

Woooo

I had a look, the maximum latency between 8pm-12am spikes well over 100ms in some graphs, but will let you know on the 31st of May on my results..

In reply to a post by narz:

Would you stress more importance on the maximum latency or the average latency?

If average latency stay low then it indicates the jitter is occasional. If maximum and average go high then the jitter is more frequent. But either way, yellow indicates jitter when green (minimum) stays low.

Given that these Plusnet graphs have shows the same patterns since Dec/Jan it would appear it's the norm for 40ms jitter in peak times and I don't think Plusnet consider that "outside design specifications" or a fault. The likes of Sky and BE generally show green with no or low yellow or blue (i.e. no/low jitter), 24/7, assuming an idle line.

Edited by Oliver341 (Wed 22-May-13 17:55:41)

In reply to a post by Oliver341:

The likes of Sky and BE generally show green with no or low yellow or blue (i.e. no/low jitter), 24/7, assuming an idle line.

Interesting. Is it safe to conclude Sky/Be are superior? (costs aside)

In reply to a post by narz:

Interesting. Is it safe to conclude Sky/Be are superior? (costs aside)

Hard to say. You can't just go by BQM, but if I was a gamer I'd probably opt for an ISP with a flat green line 24/7 over one that isn't, all other things being equal. Sky hasn't been without issues, there have been reports of half speed fibre thoughput for some time now, which may or may not be fixed for those affected.

It is rather hard to conclude whether Sky/BE's gaming experience is "superior" based on graphs made from non-gaming packets.

First, I'd have to see a whole load of Sky/BE graphs to compare.

Then (like the point made earlier), I'd have to know whether the graphs (which are for ICMP Ping packets) match behaviour for real UDP and TCP gaming packets.

If PN queue ping packets at a lower level than gaming packets (which are in the titanium queue), then PN's real gaming behaviour is going to be better than that seen in their graphs.

Then I'd need to understand Sky & BE's queuing policy. It is undoubtedly different.

Thanks for the info lads.

i'm building a BQM graph of my current O2 line before I switch to plus.net next week and will compare results. Will come back to update!

Just realised that it is a little harder to get wireshark logs of the WAN ping packets, when the router auto-responds to these.

I guess you need a PC plugged directly into the modem, running PPPoE itself.

ICMP traffic should be (and is) in Titanium queue.

I'm mostly concerned about seeing any increases in the baseline green section of the graph. That is likely to represent some sort of contention or failure between the TBB BQM servers and the router responding. We often see a baseline rise there when the router is busy doing other stuff rather than responding to the ping, or when there is exchange level congestion. It also increased when we had a device issue earlier in the year. (lots of threads about that over on community), but that was really dramatic and obvious!

Re: the peak time lift we see in the average and maximum latency? We've had a look and we think we've worked it out. Basically a lot of our traffic is gold traffic now and when there is a lot of gold traffic going through the devices we see a very small delay in completing the pass through of the gold traffic before returning to the titanium queue. What this looks like it means is the odd ICMP packet taking 20ms or so longer than average (representing the maximum latency increase) This occasional delay is enough to push the average by a couple of ms at peak. When you combine this increase with people stressing their routers at peak (after all, this is when you use it!) you start seeing those little rises.

We've quite a few gamers here in PN on our normal Unlimited product who come and punch me in the head if they see performance issues and I've checked with them numerous times and they've never seen any issues. I don't like seeing the rises on the graphs though because they do look worse than they actually appear to be!

Either way, this stuff is my (and my teams) bread and butter. If you are experiencing performance issues, yell and we'll be all over it.

Edited by deleted (Thu 23-May-13 09:41:59)

KellyD thank you for the reassuring input. I have to agree the graphs definitely do not provide the whole picture.

My O2 line at the moment is trouble free in latency and throughput, but this is what my graph at the moment looks like!

http://www.thinkbroadband.com/ping/share/831a87a979b...

In reply to a post by narz:

My O2 line at the moment is trouble free in latency and throughput, but this is what my graph at the moment looks like!

http://www.thinkbroadband.com/ping/share/831a87a979b...

Well, hopefully it will look a heck of a lot better on Plusnet fibre

In reply to a post by KellyD:

ICMP traffic should be (and is) in Titanium queue.

I'm mostly concerned about seeing any increases in the baseline green section of the graph. That is likely to represent some sort of contention or failure between the TBB BQM servers and the router responding. We often see a baseline rise there when the router is busy doing other stuff rather than responding to the ping, or when there is exchange level congestion. It also increased when we had a device issue earlier in the year. (lots of threads about that over on community), but that was really dramatic and obvious!

Re: the peak time lift we see in the average and maximum latency? We've had a look and we think we've worked it out. Basically a lot of our traffic is gold traffic now and when there is a lot of gold traffic going through the devices we see a very small delay in completing the pass through of the gold traffic before returning to the titanium queue. What this looks like it means is the odd ICMP packet taking 20ms or so longer than average (representing the maximum latency increase) This occasional delay is enough to push the average by a couple of ms at peak. When you combine this increase with people stressing their routers at peak (after all, this is when you use it!) you start seeing those little rises.

We've quite a few gamers here in PN on our normal Unlimited product who come and punch me in the head if they see performance issues and I've checked with them numerous times and they've never seen any issues. I don't like seeing the rises on the graphs though because they do look worse than they actually appear to be!

Either way, this stuff is my (and my teams) bread and butter. If you are experiencing performance issues, yell and we'll be all over it.

From my experience of seeing this on the TBB quality meter, it 's jitter, an abnormal amount as far a general networking goes,

This is likely to be caused by a link somewhere in or outside plusnet's network approaching the stages of congestion (reaching full utilisation), where the (Base latency) Green will show an increase this is sometimes coupled with some low level packet loss ,(AKA a ping tsunami BE/02 .had plenty of them) this jitter will be visable from the end users side too, as well as the likes of F8lure/firebrick latency monitoring tools ,
Another factor is a less than max throughput being possible ( unable to max out ds/us ) or if throughput is varing by a big amount example ranging from 70mbps to 35mbps during the same times as the jitter is present,

In reply to a post by narz:

In reply to a post by Oliver341:

The likes of Sky and BE generally show green with no or low yellow or blue (i.e. no/low jitter), 24/7, assuming an idle line.

Interesting. Is it safe to conclude Sky/Be are superior? (costs aside)

Different hardware and network topology at play, so what you see will be different even between 02 retail and wholesale , with 02/BE retail, the only time the yellow jitter was present was just prior and immediately following the green lag hump appearing which was peak time congestion across parts of their network or peering transit links due to their failure to invest in sufficient capacity


, I've seen a slightly worse version of this jitter problem, (quite aa lot of blue peaks below the yellow spikes)during office hours on a Friends connection, (BE/o2 wholesale) the supplier denied there being any issues,lol ,

As for BE/O2 being Superior , Their peering transits & routing to anything outside the uk was poor, and there are lots of complaints regarding high latency jitter /whild fluctuations in latency causing issues for online gamers , then there is the BBC iplayer streaming ,(buffering) issues that they never fixed ,

Edited by tommy45 (Thu 23-May-13 16:59:41)

This is an example of an ISP side problem:

http://community.plus.net/forum/index.php?action=dla...

We had a piece of kit playing up at the time. It was putting a delay on about half of the packets running through it. You can see how it forces the green minimum latency sky high. (big green spike on the left, ignore the right)

Interestingly, we hardly got any calls complaining about it. Just the guys on the forums spotting it.

They only would notice , if it affected their browsing /streaming or gaming ,VIP and unless they also had a Quality meter or f8lure graph pinging their routers or they happened to start pinging the web site etc they where having issues with, they wouldn't be aware that it was a ISP problem,as something like that would make browsing slow or would result in a timeout,

But the jitter maybe a early warning that something is reaching near capacity or some hardware /software, is possibly faulty or isn't working as it should , I wouldn't pass it off as nothing for concern , Also on the QM you posted you can see either side of the tsunami a tiny amount of the yellow jitter, and my interpretation of that graph is that who it belongs to was more or less directly connected to the faulty kit ?

Edited by tommy45 (Thu 23-May-13 17:22:47)

In reply to a post by tommy45:

They only would notice , if it affected their browsing /streaming or gaming ,VIP and unless they also had a Quality meter or f8lure graph pinging their routers or they happened to start pinging the web site etc they where having issues with, they wouldn't be aware that it was a ISP problem,as something like that would make browsing slow or would result in a timeout,

Yep. The forum peeps noticed because they were looking at their graphs, or they were gamers/VoIPing and saw a problem and used the forums to flag it.

In reply to a post by tommy45:

But the jitter maybe a early warning that something is reaching near capacity or some hardware /software, is possibly faulty or isn't working as it should , I wouldn't pass it off as nothing for concern , Also on the QM you posted you can see either side of the tsunami a tiny amount of the yellow jitter, and my interpretation of that graph is that who it belongs to was more or less directly connected to the faulty kit ?

If I remember rightly, they did also have a fault (you can see that with the big blue spikes on the right of the graph). The jitter on either side of the green spike though was actually more down to them heavily using the connection at the time.

In reply to a post by KellyD:

In reply to a post by tommy45:

They only would notice , if it affected their browsing /streaming or gaming ,VIP and unless they also had a Quality meter or f8lure graph pinging their routers or they happened to start pinging the web site etc they where having issues with, they wouldn't be aware that it was a ISP problem,as something like that would make browsing slow or would result in a timeout,

Yep. The forum peeps noticed because they were looking at their graphs, or they were gamers/VoIPing and saw a problem and used the forums to flag it.

In reply to a post by tommy45:

But the jitter maybe a early warning that something is reaching near capacity or some hardware /software, is possibly faulty or isn't working as it should , I wouldn't pass it off as nothing for concern , Also on the QM you posted you can see either side of the tsunami a tiny amount of the yellow jitter, and my interpretation of that graph is that who it belongs to was more or less directly connected to the faulty kit ?

If I remember rightly, they did also have a fault (you can see that with the big blue spikes on the right of the graph). The jitter on either side of the green spike though was actually more down to them heavily using the connection at the time.

The yellow spiking can be very similar to useage, i agree ( but can also look different depending for each connection)

But looking at the collection of TBBQM on the plusnet forum regular pattern of the yellow/blue spikes between 7-11pm (Jitter)each week night is replicated on more or less each graph that some would say maybe is indication that something isn't right, a precursor to more serious issues that customers may see in the not too distant future if they happen to be correct

In reply to a post by tommy45:

But looking at the collection of TBBQM on the plusnet forum regular pattern of the yellow/blue spikes between 7-11pm (Jitter)each week night is replicated on more or less each graph that some would say maybe is indication that something isn't right, a precursor to more serious issues that customers may see in the not too distant future if they happen to be correct

Remember it is really easy to cause a yellow spike - one packet in 100 (ie 100 seconds) will cause a yellow peak. The presence of the blue line then tells you how often it is happening.

Most of the evening peaks show a distinct yellow peak (ie the maximum has raised a lot) with a hint of blue (ie the average hasn't increased that much), and quite often with barely a flicker of the blue.

I think that fits with Kelly's explanation - that sometimes (and occasionally) the Titanium queue isn't serviced for 20ms while the Gold queue is being dealt with.

So to say that "something isn't right" about this depends on how often it happens.

If this happens to 1% of the ping packets, you'll see a yellow peak without a blue peak. If it happens to 10% of ping packets, you'll see a yellow peak with a 2ms hint of blue. If this happens to 50% of ping packets, you see a significant lump of blue.

It is hard to judge all the time, as the picture changes if you are actually downloading a lot over your own link. Therefore you can't always see anomalies on other people's graphs correctly...

However, I can say that on *my* graph (where I know very little happened yesterday), it looks like the ping packets did get these 20ms delays, but for less than 5% most of the time and 10% sometimes.

It does seem a strange queuing policy though - that you real-time titanium traffic *can* be delayed by large volumes of gold traffic. Perhaps the 20ms time-slicing is a little too coarse - even if it doesn't happen all the time.

In reply to a post by WWWombat:

It does seem a strange queuing policy though - that you real-time titanium traffic *can* be delayed by large volumes of gold traffic. Perhaps the 20ms time-slicing is a little too coarse - even if it doesn't happen all the time.

And the fact that it appears on Nearly every graph, across most if not all the plusnet network , it's the underlying cause if such exists that could be key to why this is happening, and should this be coupled with a lower level of throughput ,it would reinforce the theory of some links reaching full or near to full utilisation (congestion) or something else that isnt able to cope with the peak time traffic flow properly

Edited by tommy45 (Thu 23-May-13 20:30:26)

Very happy with my PlusNet fibre after 1 week in !!

[IMG]http://www.speedtest.net/result/2727191904.png[/IMG]

Cheers

In reply to a post by tommy45:

it's the underlying cause if such exists that could be key to why this is happening, and should this be coupled with a lower level of throughput ,it would reinforce the theory of some links reaching full or near to full utilisation (congestion) or something else that isnt able to cope with the peak time traffic flow properly

It's probably a side-effect of their traffic management/queuing systems, but 40ms of jitter at peak times isn't a bad enough issue (to them) to warrant a full-scale overhaul of their network. Maybe if the customer base grew and regular peak jitter rose to 80ms or 100ms action would be taken. We do have to consider that Plusnet are a budget ISP, whereas this would be considered a fault in need of a fix at BE (of old?).

Interesting that ICMP is in the titanium queue and still suffers the lag.

In reply to a post by Oliver341:

It's probably a side-effect of their traffic management/queuing systems, but 40ms of jitter at peak times isn't a bad enough issue (to them) to warrant a full-scale overhaul of their network. Maybe if the customer base grew and regular peak jitter rose to 80ms or 100ms action would be taken.

But reading through the Plusnet forums there are complaints about lower than normal throughput during peak times, if the two are linked then i would be suprised if it was soley down to there traffic management,

In reply to a post by tommy45:

But reading through the Plusnet forums there are complaints about lower than normal throughput during peak times, if the two are linked then i would be suprised if it was soley down to there traffic management,

I would expect throughput-affecting latency to result in an elevated green line (e.g. congestion), but this latency issue seems not to do that.

40ms is a lot of jitter.

however plusnet staff seem to be stating it isnt affecting tcp and actual gaming packets so as such isnt a major problem.

be aware if average rises to 40ms that isnt 40ms jitter. Whilst the plusnet graphs look dodgy I suspect the jitter isnt as high as 40ms on them.

what does pingtest.net report at peak?

There were complaints in Feb. That was a config issue which was resolved and unconnected to the TM platform.

If we started seeing performance issues due to kit/config, we'd be all over it to make sure we knew what was going on and trying to fix. Our reporting suites aren't highlighting anything to be concerned about at the minute.

This is why online communities are great. They help us spots issues fast.

In reply to a post by Chrysalis:

be aware if average rises to 40ms that isnt 40ms jitter. Whilst the plusnet graphs look dodgy I suspect the jitter isnt as high as 40ms on them.

As I understand it, jitter is the difference between minimum ping and maximum ping. Typically, the Plusnet graphs show a minimum ping of 20ms with maximum ping reaching 60ms ping at peak. 60ms - 20ms = 40ms jitter.

Wikipedia defines it

In the context of computer networks, the term jitter is often used as a measure of the variability over time of the packet latency across a network. A network with constant latency has no variation (or jitter).[3] Packet jitter is expressed as an average of the deviation from the network mean latency. However, for this use, the term is imprecise. The standards-based term is packet delay variation (PDV).

That implies "jitter" is the average, not the maximum.

In an example of a base latency of 10ms, with 1 packet getting a 40ms delay followed by 99 packets with zero delay: The mean latency is 10.4ms, and the total deviation is 79.2ms - so the jitter is 0.8ms

It also says that the term is imprecise, but I think they mean that people use different definitions.

I wonder how pingtest.net calculate it. They say:

"Once you understand ping, jitter should also make sense. Jitter is merely the variance in measuring successive ping tests. Zero jitter means the results were exactly the same every time, and anything above zero is the amount by which they varied. Like the other quality measurements, a lower jitter value is better. And while some jitter should be expected over the Internet, having it be a small fraction of the ping result is ideal."

I suppose that doesn't really explain their method of calculation though.

no it isnt.

eg.

10
10
10
80
10
10
10
10
10

is not 70ms jitter.

In reply to a post by Chrysalis:

no it isnt.

eg.

10
10
10
80
10
10
10
10
10

is not 70ms jitter.

Ok. What would be the jitter in your example?

You are correct.

I was not aware the type of router used will significantly affect the BQM graphs.

The O2 TG582n produces a much cleaner graph compared to TG585 v7 at idle.

TG585 v7 - firmware 8.2.7.7 - 25/05/13
http://www.thinkbroadband.com/ping/share/851136b3a58...

TG582n - firmware 8.4.4.J - 26/05/13
http://www.thinkbroadband.com/ping/share/cc9d45f493f...

Both routers work fine otherwise.

In reply to a post by Oliver341:

In reply to a post by Chrysalis:

no it isnt.

eg.

10
10
10
80
10
10
10
10
10

is not 70ms jitter.

Ok. What would be the jitter in your example?

I think they would yield 21-22ms jitter.

17.5 actually.

I'd just like to add to this: my Samknows monitor shows little (if any) corresponding increase in 'RTP Jitter' during peak evening times. Yesterday's graph, for example, shows a variation from 0.4 to 0.7ms between 4PM and 11PM where the TBBQM shows a blue peak of around 3ms betwen 9:30PM and 10:00PM, conversely the Samknows graph shows a single peak of 2.45ms at 2PM with no visible blue on the QM around this time.