Latency spikes

Hi

I have FTTC approx 50/10. Every test available shows an average ping between 7-15ms. According to the monitor, I have almost constant latency spikes between 80-140ms. The buffer bloat results seem to yo-yo accordingly.

I have had constant issues with shots registering in online games for years and believe this to be the reason. I have some sort of [smart/adaptive?] QOS enabled on my relatively expensive ASUS router/modem. The model number escapes me just now. Anyway, I�m hoping that eliminates buffer bloat on my end.

So I guess my question is, where would the slow down be more likely to be occurring, Plusnet, or my local exchange? The latter being 3 miles away in a heavily built up area.

If anybody could post an image of the monitor with a [real world] example of stable latency, that would help me manage my expectations while recording my own over the next few weeks.

Thanks.

https://www.thinkbroadband.com/faq/broadband-quality...

Scroll down that page to a series of BQM graphs and there is a nice almost all green one.

Thanks.

So my graph looks similar to the �normal connection� graph, except the yellow spikes climb about 50% higher (average spike being 120-140ms).

Does the monitor show the effect of buffer bloat, as recorded by the TBB speed test? I perhaps wrongly assumed that was the reason for the spikes. It�s difficuot to discern the width [duration] of the spike to a short enough timescale. Does the steady average latency suggest that they must be repeatedly short?

Please could you link to your BQM, and what router do you have?

Some are known to give a regular spike, easily recognised, but I can't remember which routers do.

Link to the BQM is best at this stage

Each pixel is 100 seconds, so yellow may be just 1 or 2 samples (1/second) having higher latency, i.e. if the thin blue line does not move then what you have is jitter and this can be caused by simply loading a web page.

https://www.thinkbroadband.com/broadband/monitoring/... is my actual connection from today, the yellow spikes showing when I was busy using the connection, web pages/email/small files today

The long red spike at 2:15pm was a reboot of the router i.e. I pulled the power
Big yellow spike at 7:45pm is me uploading an image to twitter

This is the monitor.

This is the router.

So get up around 6am and household is quiet from 8am to 3:30pm when some gets home from school?

Then busy streaming video or other activities from 8pm and still ongoing now

To know if you have a jitter/latency problem you need to stop using the connection for 30 minutes or so and see when the BQM is still showing variations

Ok will do.

Live graph.

The major batches of yellow are typical of something that is being done on computer equipment in the home, possibly not by yourself though obviously some of it will be.

I doubt if my suggestion of a regular spike applies, as my ASUS RT-N66U doesn't exhibit them. Though there is a possible pattern of regular spikes to about 25ms, with some lost inside the wider patches. Starting with the 42ms just after 2am, then just before 5am, 9:30am; 12:20; just before 3pm, then obliterated.

Whatever problems you are encountering in real use, this is nothing to do with it even if your different model router is doing this spiking.

I hope MrSaffron manages to spot something.

In reply to a post by MrSaffron:

So get up around 6am and household is quiet from 8am to 3:30pm when some gets home from school?

Then busy streaming video or other activities from 8pm and still ongoing now

To know if you have a jitter/latency problem you need to stop using the connection for 30 minutes or so and see when the BQM is still showing variations

Pretty much spot on.

Looking at the live graph in my previous post, I can tell you that everybody was turned off at midnight (00:03 to be precise) and you can still see spikes. Not sure what could have happened at 4.00am. I will investigate whether anybody left a console on standby, that could have updated itself in the background etc.

What I�m trying to understand is this. If I have a solid average ping of say, 15ms, and a good connection, should I not be able to sustain say, 10Mbps without spiking? Is the spiking down to contention, or as I have tended to suspect, external buffer bloat? Spikes of 120-140ms are very high to my mind, if I�m not actively viewing information in far flung locations. I can reach some servers in New York in 75ms, so I start to wonder about traffic and routing etc.

I understand that a connection can�t be �perfect�, but without a reasonable expectation as to what spikes are normal, it�s difficult to look for patterns in behaviour.

In reply to a post by mrnelster:

Not sure what could have happened at 4.00am. I will investigate whether anybody left a console on standby, that could have updated itself in the background etc.

I know Sony likes to do PS4 Updates around that time.

I might be wrong and I know people will correct me if I am, but I am wondering if it was an upload, reason being pings require very little bandwidth which is why you hardly see the blue parts in the BQM due to you are mostly using the download hence the large yellow bars, but if the upload is being used up then it has to wait to be able to send a ping response, hence the blue bars.

I know I only see large blue parts when I am uploading large files like in my BQM at 12 pm which was a 2.9 GB video file to YouTube.

But like I said I might be wrong.

But yeah it could be just an update to a PC or Console that took about 30 mins to complete.

Paul

4am does look like something work up and uploading or downloaded something e.g. mobile on Wi-Fi doing backup, app updates automatically, console coming out of sleep

Otherwise no regular pattern to speak of.

You need to have a night where router has Wi-Fi turned off and no Ethernet cables plugged in.

Latency spikes when downloading are not normally caused by slowness of data, but usually by the different priorities the routers along the route apply to data and pings. As described in the FAQs and enlarged on by Paul.

When any router between you and the tbb server is busy it has more important things to do than respond to a ping, so it sends the response when it has dealt with the data, including the ACKs and on G.INP any retransmission requests. It is only if red down-spikes at the top appear that throughout questions are getting serious as that means ping packets and maybe data packets are being dropped at some point, in either direction.

Some routers are configured never to respond to pings, as you can see on many tracerts, typically shown as timeouts on all three attempts at that hop on every run of tracert on different days.

Timeouts and anomalous high response times on tracerts, where subsequent hops are low, suggest a router working very hard. A set of high times from a particular router followed by times that are never lower than that router show a bottleneck/congestion at that point. Both cases will almost always cause data slowdowns, including your gaming packets in both directions, even if no ping packets are dropped.

In reply to a post by mrnelster:

Looking at the live graph in my previous post, I can tell you that everybody was turned off at midnight (00:03 to be precise) and you can still see spikes. Not sure what could have happened at 4.00am. I will investigate whether anybody left a console on standby, that could have updated itself in the background etc.

Also a smart TV left on standby perhaps? I also know several people who thought when they press the power button on an iPad until it clicks and the screen blanks they had turned it off. Until they mentioned the battery life was poor and I enlightened them. They were using things like multiple weather apps updating frequently, Gmail running, even Skype not closed down.

Nothing uploading intentionally at that time, but I do have a PS4 sat in rest mode. I�ll
have a look at that, as I wasn�t aware that was the sort of time it might update. Thanks for the tip-off.

You need to have a night where router has Wi-Fi turned off and no Ethernet cables plugged in.

Difficult to achieve, but I will try to plan it.

Am I right in thinking then, that the broadband monitor will not really depict buffer bloat if it is only drawn every 100 seconds? The delays only need be a fraction of that to cause my issue, though relatively frequent.

You may have mis understood each pixel may reflect 100 seconds, but as the yellow is the maximum latency any high sample in the 100 samples in that period should show up.

In reply to a post by MrSaffron:

You may have mis understood each pixel may reflect 100 seconds, but as the yellow is the maximum latency any high sample in the 100 samples in that period should show up.

So what you are trying to say, is that the monitor pings every second, and the yellow spike will be the highest of those hundred responses?

Yes

Had a look. Nothing from the PS4 last night.

Thanks, I�m with you now.

In reply to a post by mrnelster:

So what you are trying to say, is that the monitor pings every second, and the yellow spike will be the highest of those hundred responses?

And the blue is the average for that second, so also important if it gets very high.

Some minor packet loss at times last night.

Live graph.

So what am I meant to be learning from all of this; maybe I am not understanding the basics clearly?

Your monitor is poking my router and waiting for a response. If my router is set up correctly, and QOS is doing it�s job, why would loading a web page create such a spike? I always reasoned that unless my connection was being saturated (or my router overwhelmed) that the latency would remain relatively constant. Even with a couple of people streaming, YouTube/Netflix etc, in principle I didn�t envisage such increases in latency.

That�s why I bought a higher end [domestic] router. So if it isn�t queuing at the router, why such high latency unless I am saturating my bandwidth? If I�m consuming 15Mbps solidly, on a 50Mbps/10ms connection, theoretically other than traffic/contention issues, should I expect the monitor to report such elevated responses?

Am I missing something fundamental in my reasoning, or is this exactly what the erratic buffer bloat tesults are trying to tell me? In real world terms (for gaming) I would be better served by a constant 50ms ping, than being recognised as a 10 whilst regularly bouncing up to 140, if you get my drift. Prediction must be terrible.

So I think what I need to do next, is to turn everything in the house off and record a gaming session on the BQM.

Do you have QoS set up?

YouTube/Netlix may say a 4 Mbps rate but if you watch the traffic flow it is often spikey and its those spikes

Any use of the connection can impact things, unless your router has a processor dedicated to responding to ping requests

6 random pixels of packet loss - not enough to lose sleep over

Yes QoS is on and prioritising gaming. I read somewhere that latency beyond 100ms causes VoIP users to talk over one another. I often get that during in-game chat.

Next step will be everything off for a few hours. Then a gaming session with everything else off.

But does the QoS prioritise your voice chat? Gaming QoS settings probably just deal with gameplay rather than the voice comms

Or worse the QoS is not doing what you expect at all, i.e. different games will behave differently

If nothing else using the connection you can test if QoS is helping or not, by playing with it off and with it on. Also make sure you are using a nearby server when evaluating gaming since there may be lots network between you and server and may be that rather your broadband that is the issue

In reply to a post by MrSaffron:

But does the QoS prioritise your voice chat? Gaming QoS settings probably just deal with gameplay rather than the voice comms

Don't know. Will look into that, but that issue pre dates this router I think.

Or worse the QoS is not doing what you expect at all, i.e. different games will behave differently

If nothing else using the connection you can test if QoS is helping or not, by playing with it off and with it on. Also make sure you are using a nearby server when evaluating gaming since there may be lots network between you and server and may be that rather your broadband that is the issue

That is essentially what I am trying to establish. What effect the buffer bloat may be having on my connection. If because of it the matchmaking servers are lumping my connection in with consistently poorer quality ones, that matters.

Out of interest, while reading up on buffer bloat I found fast.com recommended as now measuring the effect of load on latency. The test shows my exhange as different to the one I have been connected to historically; the same one TBB and SamKnows has always agreed with.

Does anybody know if the FTTC can be run in from a different exchange, or could the original exchange be linked to the more central one that fast.com is stating?

My exchange has always been Stny Stratford - fast.com shows it as Fishermead.

How does fast.com show your exchange?

It reports a GeoLocation based on your IP address which is NOTHING to do with the exchange and can be hundreds of miles out from your real location.

Not at all sure on its latency measurements during upload phase, as it reports 400ms whereas I see around 100ms from our tester. Looking at network traffic I think it is actually reporting the latency for the upload packets rather than attempting to download a small packet in the middle of the upload stream.

Dunno. Just says Fishermead, which is actually the Bradwell Abbey exchange and the main one in Milton Keynes. We are closer to it than the Stony Stratford exchange we have historically been connected to. It says �Client: Fishermead� so wondered if that meant we were passing through it in some way.

I know geo location can be anywhere, but interestingly mine is at the opposite end of the city. So connected at one end of the city, geo location at the other end, and the �Client� smack bang in the middle. Seems a pretty daft way of going about things to me. Misleading, pointless information.

I�ve never been to fast.com before. I clicked it after visiting bufferbloat.net

Haven�t read it properly yet or tried their quick test, as I haven�t been near my PC, but it looks interesting.

They use geolocation I know, because puts me in Hitchin where IDNet are based

Geolocation via IP address is no good for identifying locations in the United Kingdom, plain and simple so ignore that field on fast.com

I often get the adverts on advert-carrying websites that say "Laser eye surgery" or some other service is going viral in my location. It used to say Stockport when I was with Plusnet. These days it is usually Arnold! That's near Nottingham and AAISP HQ is in Bracknell.

Just had a thought. I must watch out for these ads to see where it says now my base latency has risen.

In reply to a post by MrSaffron:

Geolocation via IP address is no good for identifying locations in the United Kingdom, plain and simple so ignore that field on fast.com

For fun, try this page: https://www.iplocation.net/

I like the understatement �The US domain names such as .com, .net and .org does not always imply that the host is located in the United States�.

In reply to a post by jchamier:

For fun, try this page: https://www.iplocation.net/

That was funny, I just check my connection and I was all over the place and it was way off, I am located in East London (about several miles from the London Eye) LOL

IP Location Fun Stuff 001.png

Paul

Yes IP location doesn�t work in the UK unless you are on Virgin Media�s cable network. Even on VM you tend to find the neighbouring large town, rather than your own town.

I�m still interested in what this test is [really] telling me. So I have a gaming QoS set up. The family are fiddling about throughout the evening on Netflix/Youtube etc. What does the ping spike usefully show me? To my mind, and ignoring for a moment what might be an �average� graph, my expectation from the ISP is the narrowest margin possible between loaded and unloaded latency.

So the question is, with QoS set up, does the monitor ping get sent to the back of the queue? I would assume that an excellent connection graph should look bumpy, not tall and spikey. I understand that real world scenarios aren�t perfect, but I don�t think constant spiking denotes a quality connection, even when I�m draining a constant 10Mbps down.

I�m not talking about real world, just theoretically what a graph with a 7-10ms latency would look like, alternating between loaded and unloaded as my usage scenario above?

Without having your exact QoS to test on a known well behaving connection impossible to say...

Some QoS systems can show perfect latency for ICMP but if a game is using some TCP packets they may not be QoS preferrenced

i.e. almost all the simple turn on QoS systems are simple things like prioritise ICMP or UDP packets over TCP, or reserve 10% of bandwidth for a certain activity. In short you need to experiment with settings of your QoS system to determine what works best for you.