PDA

View Full Version : packet loss? tcp issues? I've got no idea.


randomdude
25-02-2007, 21:53
OK, I'm confused here.

Back story is that I live in a 5-person student house. I'm in charge of the net access. A couple (or three) of the guys in the house torrent fairly heavily. We have a Telewest cable line, live somewhere around Leeds, and use a surfBoard cable modem. Haven't called tech support. Home network setup is a linux router but this was unplugged for most of the testing (see below text) and I was mainly using an XP laptop and another XP box for testing. No firewalls are in effect other than the Linux machine.

A few weeks ago we started having network congestion (2+ minutes to load a web page, etc). This has steadily been getting worse, and recently I hassled the torrenters and eventually removed them from the network, which eased the problem, but didn't get rid of it totally. I got annoyed tonight and unplugged the lot, jacked my Xp laptop into the cable modem and rebooted it. I get around 5% packet loss pinging bbc.co.uk (is this an acceptable level?), but I still take over 30 seconds to load, for example, my LJ friends page, although most sites are responsive. I tried a nearby other machine (also running XP) as well.
So I do a packet dump and find (a substantial amount of) 'TCP Retransmission' and 'DUP ACK' packets. The page takes 30 seconds to load. netstat tells me I have had 8% packet loss during the http get.
This makes me suspect packet loss, so I go to my cable modem's stats page and it tells me my snr is 39dB, downstream power level is 5dB and the upstream power level is 39dBmV. This is in line with what I've read on the subject. I've unplugged everything else (cable box, reception is OK) from the coax line.
Now I'm really at a loss. Surely any problem that 'isn't my problem' would be reflected in the cable modem stats? Or is it likely that my modem has snuffed it? Or have I/my housemates annoyed someone at Telewest with their torrenting, who have promptly banhammered us, causing what I see?
I'm really confused. I've got all my housemates screaming at me to 'fix it' and I'm wondering if its time to call support - but don't really know what to say and really dont want to get an engineer out only to find its my problem and not his.
Any pointers, ideas, or anything? I'm all out of ideas :(

You say you like bbc.co.uk traceroutes, so here's one.

C:\Documents and Settings\a>tracert bbc.co.uk

Tracing route to bbc.co.uk [212.58.224.131]
over a maximum of 30 hops:

1 <1 ms 2 ms <1 ms 10.xx.xx.xx
2 13 ms 21 ms <1 ms 80.xx.xx.xx
3 <1 ms 1 ms 1 ms 62.30.252.26
4 1 ms 1 ms 1 ms 194.117.136.166
5 3 ms 1 ms 1 ms 194.117.136.173
6 <1 ms 1 ms 1 ms 194.117.136.146
7 2 ms 1 ms 1 ms bbc-gw0-linx.prt0.thdoe.bbc.co.uk [195.66.224.103]
8 2 ms 2 ms 10 ms 212.58.238.153
9 5 ms 3 ms 3 ms rdirwww-vip.thdo.bbc.co.uk [212.58.224.131]

Trace complete.

This is one of the things that confuses me - all that gets through fine, as does the pinging (with slight packetloss) but a TCP sesh really breaks things.
As I say, I'm all out of ideas. Can anyone see anything I've missed, or any issues I should phone Telewest about? I suppose a good deal of what I'm asking can be summed up as 'is this all normal'.
Thanks.

bigcat
26-02-2007, 14:11
Have you tried the obvious?

A virus scan & spyware scan?

Carl J
26-02-2007, 14:46
That's a very very odd traceroute indeed, those pings aren't achievable over cable.

Can you do a trace from the Linux box perhaps?

5% packet loss is entirely unacceptable, packet loss should be near as damnit zero. 0.05% packet loss might be ok :)

randomdude
26-02-2007, 20:01
That's a very very odd traceroute indeed, those pings aren't achievable over cable.

Can you do a trace from the Linux box perhaps?

5% packet loss is entirely unacceptable, packet loss should be near as damnit zero. 0.05% packet loss might be ok :)

I'm glad someone agrees with the packet loss being unacceptable - I wasn't sure if it was or not :)
Re the traceroute, though, that *is* odd. I didn't pick up on that. Re-running today on that machine gives all figures as '<1ms'.. a different windows box gives more sensible results. Perhaps that machine is hosed, then..

Z:>tracert bbc.co.uk

Tracing route to bbc.co.uk [212.58.224.131]
over a maximum of 30 hops:

1 2 ms <1 ms <1 ms 192.168.30.254
2 41 ms 18 ms 35 ms 10.153.0.1
3 15 ms 10 ms 22 ms 80.194.0.37
4 23 ms 27 ms 24 ms e123-van1-kno.cableinet.net [194.117.138.2]
5 20 ms 19 ms 23 ms 194.117.136.149
6 * * * Request timed out.
7 258 ms 188 ms 245 ms 194.117.136.146
8 24 ms 43 ms 51 ms bbc-gw0-linx.prt0.thdoe.bbc.co.uk [195.66.224.10
3]
9 31 ms 16 ms 15 ms 212.58.238.129
10 20 ms 23 ms 36 ms rdirwww-vip.thdo.bbc.co.uk [212.58.224.131]

Trace complete.

odd that hop 7 is so slow, and the things behind it aren't..
I'm still out of ideas. Perhaps I should unplug the LAN from the modem again, and plug it into a different machine (ie, not the one that gives sub-ms pings) then do a packet capture of some http and some traceroutes.
Incidentally, everything appears to be working fine now that I've plugged everything back in and told everyone in the house 'no torrenting for the next few days'.. it seems that the packet loss is related to the load (well, obviously).
I'm gonna try a heavily throttled torrent and see if everything breaks. Wish me luck...

---------- Post added at 19:01 ---------- Previous post was at 18:14 ----------

Right, I've re-tested things from another machine - one that gives sensible traceroute timings.
When I reload my LJ friends page (yes yes I know), sometimes the page will load with no problems, or sometimes I get what appears to be one packet lost ('previous segment lost', 'Duplicate ACK', then 'Fast retransmission'). This kinda worries me - that if I call an engineer and the problem vanishes, I feel a bit silly. The page is also loading at a proper speed now - about 5-10 seconds for images and all.

So basically, it all magically works (from all machines, including those that didn't work yesterday). :/

Chrysalis
27-02-2007, 00:49
congested ubr by looks of that tracert

randomdude
28-02-2007, 18:20
congested ubr by looks of that tracert

That'd certainly tally with what I'm seeing. I think things are breaking mainly during eveningtime peak hours. What do I do about it, though? Phone support and tell them 'it stops working after about 6pm'?

jerryl
28-02-2007, 19:46
i think you can phone 0871 225 1111, this tells you information on whats happening like if theres networking maintenenace in certain areas and when its going to stop, if theres some networking going on in your area then maybe they r trying to fix ur congested UBR problem, ive got a congested area, so i get high pings alot and slow internet, i dialed that number up and it said networking maintenance is happenng between 22nd to 29th of february.
Maybe you just have to wait until VM fixes it ,they probably know the problem already.
I would ring up customer service and ask if there is anything going on in your area.

randomdude
28-02-2007, 20:06
OK, thanks a lot for the advice and help then, everyone :)
Forums full of random knowledgeable people helping all-comers is what keeps the Internet spirit alive. You guys rule :)

-Randomdude / Alan