Anyone else seeing dodgy throughput

[andrue]

Yah, getting worse here. TBB is badly affected, I'm barely getting over 5Mb/s. Speedtest.net is down to 20Mb/s. Both highly variable. I've noticed this on other occasions - the link to TBB often suffers more.

[sobranie]

Other ISP users around my area seem unaffected. Is it reasonable to conclude that all would be Apple Site perusers are availing themselves of IDNets bandwidth/servers etc etc.?
FWIW my speed is now sub 4.
Would it also be reasonable at this juncture for IDNet to consider any other scenarios to explain this diabolical situation?

[Steve]

I'm reminded of this post

http://www.idnetters.co.uk/forums/index.php/topic,31394.msg718609.html#msg718609

It does seem disappointing that the peering links are being blamed for our current poor throughput,I'll take your word that other ISPs are not suffering, certainly it's after LINX peak time and my throughput is still poor.

[JamesAllen]

Yeah looks like it got better about 19:45 but then from 20:00 has been bad again. This is really annoying. Hoping I can online game later but if the packet loss continues it won't be worth it.

[Gary]

Considering no other ISP's are mentioning this...and the fact my mother on BT infinity is getting a happy 38Mbps down at this very moment it feels like an issue with IDNets peering points, and very ironic after that post Steve mentioned. 

[andrue]

I'll take your word that other ISPs are not suffering, certainly it's after LINX peak time and my throughput is still poor.

No need to take it on trust. Look here:

http://craigswebsites.co.uk/ping/

Looks to be fairly unique to IDNet although there's one Supanet monitor showing packet loss of around the same duration.

[syserr0r]

I'm reminded of this post

http://www.idnetters.co.uk/forums/index.php/topic,31394.msg718609.html#msg718609

It does seem disappointing that the peering links are being blamed for our current poor throughput,I'll take your word that other ISPs are not suffering, certainly it's after LINX peak time and my throughput is still poor.

It would appear that post was designed to be misinterpreted.

My understanding (supported by information from here) is that they have a handful of uplinks (In this case the main uplinks appear to be 1xLINX Gigabit link and 2xLONAP Gigabit links [with both IPv4 and IPv6 respectively]) that give IDNet their upstream connection to the 'rest of the internet' (aka the 600 other peers that also connect to LINX/LONAP; and via LINX/LONAP to connect to Telia/Level3 etc.)

A bit like boasting that my phone can connect with over 6 billion different mobile phone numbers, when that applies to anyone with a mobile phone. (And incidentally, I can't as soon as my mobile phone loses coverage).

[Steve]

To be honest Gary I wouldn't be bothered at all, we get the odd blip and that's life,it just coincides with my desire to download what seems the entire contents of Apple's server.

@syserr0r I wonder how throughput is currently on IPv6

[andrue]

To be honest Gary I wouldn't be bothered at all, we get the odd blip and that's life

Yeah, I don't currently see it as a problem because I'm sure it will soon clear. My main concern last night was if it was something local to me.

[Gary]

To be honest Gary I wouldn't be bothered at all, we get the odd blip and that's life,it just coincides with my desire to download what seems the entire contents of Apple's server.

@syserr0r I wonder how throughput is currently on IPv6

Be bothered with what, Steve?  Twelve hours of sub par speeds is not a blip...

[JamesAllen]

It certainly isn't. Well packet loss is still awful so looks like online gaming is a no no tonight. Really annoying as this is one week I get off from planning my radio show. Sods law...

[Steve]

Be bothered with what, Steve?  Twelve hours of sub par speeds is not a blip...

I'm speaking for myself, life goes on, in the whole scheme of things it's not that important to me. If it becomes a regular event then it would impinge on me.

[syserr0r]

@syserr0r I wonder how throughput is currently on IPv6

Likely wouldn't be any different. I would guess that both IPv4 and IPv6 traffic is routed over the 3 main (Gigabit) peers (although they could be separate¹), so congestion would affect both.

It has really surprised me to find they rely on 3Gbit/sec (and 350Mbit/sec if you're being picky) for their uplinks.
I didn't imagine them to be a big company, but some rudimentary sums give, assuming:

Download

• IDNet Avaiable bandwidth: 3,350 Mbit/sec
• Typical FTTC Bandwidth*: 40Mbit/sec
• Typical ADSL Bandwidth*: 10Mbit/sec

*Assuming 50% of Max. theoretical throughput is the average

Would allow for:

• 3,350 / 40 = 83 FTTC Users at maximum download
• 3,350 / 10 = 335 ADSL Users at maximum download
• 3,350 / 50 = 134 Equal-split of FTTC/ADSL Users at maximum download

At any given time


Upload

• IDNet Avaiable bandwidth**: 3,350 Mbit/sec
• Typical FTTC Bandwidth**: 10Mbit/sec
• Typical ADSL Bandwidth**: 0.75Mbit/sec

*Assuming Full-Duplex uplinks (the almost certainly would be)
**Assuming 50% of Max. theoretical throughput is the average

Would allow for:

• 3,350 / 10 = 335 FTTC Users at maximum upload
• 3,350 / 0.75 = 4,466 ADSL Users at maximum upload
• 3,350 / 10.75 = 622 Equal-split of FTTC/ADSL Users at maximum upload

At any given time

The scary thing is it wouldn't take more than (3,350/72 [80Mbit minus 10% overhead] = ) 47 FTTC users with maximum line speed to completely saturate their uplinks.

Ah, the joys of playing the over-provisioning game!

¹Double the number of users if the IPv4 and IPv6 uplinks are separate (so 94 max-speed FTTC users to saturate all their uplinks, or 47 max-speed FTTC users to saturate only ipv4 or only ipv6)
For anyone interested, take a look at TalkTalk's uplinks

[Steve]

I guess if your assumptions are correct it's quite feasible to see how congestion will occur.

[Gary]

Well I'm back to full speed again,hope it lasts 

[psp83]

Its getting better but speeds are still all over the place.

[Fizzy]

I seem to have an improvement now - speedtest.net just about back to normal.   thinkbroadband tester giving odd results again though :-



I'm confused about the TBB test.  If IDNet don't do any form of packet shaping or traffic prioritisation, how come there's such a difference?  Surely it should be the same for both, or both equally effected by network traffic....although I guess it depends on where the target servers are that the tests are using too... 

I had an issue on my pc before where the TBB test was low but http full speed.  Was fine on a laptop using same cable/router though.   Ended up rebuilding my pc as it seemed to be some wierd networking issue.  Anyhow, this time it's not machine specific.   

[Steve]

Certainly my packet loss seemed to resolve at around 11pm some 14 hours after it started.

[Bill]

LONAP traffic for the last 48 hours- comparing the shapes of yesterday's curve with the day before (and ignoring the Mavericks spike) it looks as though from around 10:30 to 15:00, from 16:00 to 18:00 and (maybe) 19:30 onwards a fairly hefty chunk of traffic was routed somewhere else, then it returned to normal (ie matched the previous days pattern) at around 22:00. Which (broadly) fits with my BQMs.

I'm reluctant to believe that those drops are due to straightforward lack of internet usage... whether it was routine maintenance, equipment failure or something else completely, and whether IDNet traffic was moved or stayed on the old path, I have no idea but it seems likely to me that there's a connection somewhere.

Anyone more knowledgeable think that seems reasonable?

[Gary]

Interesting Bill, I was in the middle of downloading when I lost throughput at 10:30am ish  I have called a few gaming friends and none seem to have seen the packet loss we experienced. 

[Simon_idnet]

It appears that we have a large number of Apple fans who like to update the software on their devices promptly. As none of our LLU customers experienced any problems it seems that we'll need to get some more bandwidth out of BT. Apologies for the inconvenience.

Bill, I see that Akamai (who host content for Apple and BBC etc) moved their traffic away from LONAP and over to LINX and other peering points a few times over the past few days.

[Bill]

I wasn't completely wide of the mark then, thanks Simon

From my BQMs it looks like they're doing it again, but a bit early to be sure yet.

[Gary]

Apple maybe should learn to release updates in a staggered manner rather than everything all at once 

[Bill]

Or users develop some patience instead of wanting it NOW

[Gary]

Or users develop some patience instead of wanting it NOW

I think Apple releasing updates on certain days once a month is probably a more realistic approach