[TelstraClear] TelstraClear Cable connections appear to be rate shaped in one way or another

Forums › One New Zealand (including Vodafone, WxC, Farmside) › [TelstraClear] TelstraClear Cable connections appear to be rate shaped in one way or another

1 | ... | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12

176 posts

Master Geek

#526453 27-Sep-2011 20:30

sbiddle: You of course keep making the assumption that something is broken. What if it's not?

No one (official) has told me nothing is broken, and Telstra are looking into it. I'd assume they wouldn't look into something they knew wasn't broken.

If it's not broken, I switch providers.

jpollock

600 posts

Ultimate Geek

Trusted

#526475 27-Sep-2011 21:05

sbiddle:

You of course keep making the assumption that something is broken. What if it's not?

Then we learn something really cool about TCP, specifically how the window size extension doesn't work. That would be a pretty interesting blog post, and likely to get some hits from HackerNews, Slashdot and ArsTechnica.

Jason

---
http://blog.jason.pollock.ca

jnawk

176 posts

Master Geek

#526588 28-Sep-2011 07:48

This just in: I tried a download from my server in Germany this morning - instead of the usual steady crap rubbish speed of around 230-280kb/sec (1.8-2.2mbit/sec), it did, for a time, burst up to 900kb/sec - 7.2mbit/sec.

I also did a speedtest.net - to Dreamhost in LA - got 8mbit/sec - during my download from Germany, itself pulling around 3.5mbit/sec. That result is consistent with what I got on the 6th June, but way faster than what I got on the 17th of September (4.03mbit/sec)

So, I'm thinking there is a problem, and Telstra is being mum about it, that's why I've not really heard back from them since Friday.
If that's the case, I have one thing to say: Keep up the good work - it looks like you're nearly there!

maverick

3594 posts

Uber Geek

Trusted

WorldxChange

#526622 28-Sep-2011 09:21

Talkiet:
jnawk:
Talkiet: I'd also caution you about switching if you expect full rate TCP throughput on a single thread.

XNet can do it. What is different in their setup?

I didn't say the Telecom BB connection wouldn't do it... I cautioned you that it's not explicitly designed to do it. It _MIGHT_ deliver it. We test throughput from ADSL connections but if we see less than line rate for a single thread connection to a random server outside of NZ we don't mark it down as a fail!.

If you know that XNet will reliably deliver this capability, then I suggest they should be your first choice for a replacement provider.

(evil grin at any WXC staff watching :-)

Cheers - N

You know us little TinPot companies, we do some things okay Neil .... Sly Grin

Yes I am a employee of WxC (My Profile) ... but I do have my own opinions as well

https://www.facebook.com/wxccommunications

PenultimateHop

637 posts

Ultimate Geek

Trusted

#527034 29-Sep-2011 06:36

jnawk: If the capacity was not there, I'd not be able to get 15mbit/sec. That I can get 15mbit/sec (I can) clearly indicates the capacity is there.

It indicates the TOTAL capacity is there, but it does not indicate that individual flow capacity is there. If there's any form of multi-pathing (ECMP, LAG, MLPPP - possibly even in DOCSIS3.0 channel bonding, but I'm sure) this can easily not be the case.

jnawk

176 posts

Master Geek

#527039 29-Sep-2011 07:01

PenultimateHop:
jnawk: If the capacity was not there, I'd not be able to get 15mbit/sec. That I can get 15mbit/sec (I can) clearly indicates the capacity is there.

It indicates the TOTAL capacity is there, but it does not indicate that individual flow capacity is there. If there's any form of multi-pathing (ECMP, LAG, MLPPP - possibly even in DOCSIS3.0 channel bonding, but I'm sure) this can easily not be the case.

This is all starting to look very unlikely, isn't it? We're talking about 15mbit/sec through ISP's networks, not some leased line to a satellite office.

We've confirmed 10mbit/sec is there (other dude's connection), but anyway, why should the core care about flows? That's a TCP concept, and all the routers in between "see" are IP packets.

PenultimateHop

637 posts

Ultimate Geek

Trusted

#527042 29-Sep-2011 07:21

They hash accordingly, using some consistent criteria (often L4 due to lack of entropy at L3).

Equinox IT

Cloud spending continues to surge globally, but most organisations haven’t made the changes necessary to maximise the value and cost-efficiency benefits of their cloud investments. Download the whitepaper From Overspend to Advantage now.

jpollock

600 posts

Ultimate Geek

Trusted

#527627 30-Sep-2011 12:04

Another Friday, any updates from TCL?

---
http://blog.jason.pollock.ca

jnawk

176 posts

Master Geek

#527960 1-Oct-2011 09:06

jpollock: Another Friday, any updates from TCL?

My other friend with the Telstra connection is having performance issues. He quoted the ticket number to the helpdesk, and they told him the issue was resolved. Like hell!

Perhaps I should have Gary up on that.

But I'm on holiday till Tuesday, so don't really want to get all worked up about it yet :-)

jnawk

176 posts

Master Geek

#528919 4-Oct-2011 11:15

Telstra Tech emailed me today - issue is still being looked at.

mercutio

1392 posts

Uber Geek

#529713 5-Oct-2011 17:27

TelstraClear have had slow international performance for ages. When I tried investigating I noticed that most sites showed low grade continuous packet loss irrelevant of stream speed.

TCP/IP is horrible for dropping speeds at even low amounts of packet loss.

jnawk

176 posts

Master Geek

#529724 5-Oct-2011 18:06

mercutio: TelstraClear have had slow international performance for ages. When I tried investigating I noticed that most sites showed low grade continuous packet loss irrelevant of stream speed.

TCP/IP is horrible for dropping speeds at even low amounts of packet loss.

That would explain it. Incidentally, dropped packets are one of the mechanisms of implementing shaping...

mercutio

1392 posts

Uber Geek

#529726 5-Oct-2011 18:10

jnawk:
mercutio: TelstraClear have had slow international performance for ages. When I tried investigating I noticed that most sites showed low grade continuous packet loss irrelevant of stream speed.

TCP/IP is horrible for dropping speeds at even low amounts of packet loss.

That would explain it. Incidentally, dropped packets are one of the mechanisms of implementing shaping...

Well I would say that it's just their links being congested but it doesn't seem to be that related to time of day, and it seems to happen to any2ix etc.

That said I remember a while ago pacnet had bad performance at any2ix...

(any2ix and equinix are the two big los angeles internet exchanges)

Pretty much where I've settled now days is that if there's more than 0.5% packet loss it's sufficient to complain about.

That said, even with really small amounts of packet loss dropped in UDP, TelstraClear seems to give slower performance with TCP/IP. Which could in theory be due to TCP/IP bursting a whole lot of packets in a short period of time.

Like if you're sending at 500k/sec. then tcp/ip may send 50k in 1msec, then no traffic for 99 msec, then 50k .. then no traffic ..

There is some level of packing by the acknowledging of data increasing the window size one packet at a time.. but there's still nothing like even distribution of packets.

And let's say you have a buffer of 15 packets. And 16 packets get sent in one lump by tcp/ip, then the last packet will get dropped.

Although now days, most web servers etc seem to be connected at gigabit speeds, and most infrastructure at ten gigabit .. the situation would actually be worse if around the other way..

Still, with cable modems etc .. they have to be able to queue packets before it gets sent on to the cable modem because packets come in faster then they can be sent to the end user ..

Then if you queue lots of packets, then you get latency spikes. So you need multiple queues for different traffic .. which is complicated and expensive. So often there's quite simple algorithams like RED ...

http://ieeexplore.ieee.org/Xplore/login.jsp?url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F5462646%2F5469698%2F05469800.pdf%3Farnumber%3D5469800&authDecision=-203

Although that problem exists in different ways on both upstream and downstream .. last I knew cable modems were congesting when doing a single upload all the time. But downstream wasn't congesting to the same degree even factoring in the speed difference suggesting a smaller queue size. Like an upload may congest your connection by up to 400 msec with 2mbit upload 10 mit download.. but a download congests your connection by up to 20 msec ..

jpollock

600 posts

Ultimate Geek

Trusted

#540581 3-Nov-2011 08:53

Any updates on this?

---
http://blog.jason.pollock.ca

Ragnor

8219 posts

Uber Geek

Trusted

#540708 3-Nov-2011 13:22

jpollock: Any updates on this?

I recall a post on GZ in a different thread from the dude on Snap VDSL2 (Snap use Telstraclear/Reach for part of their international transit) who also identified the same issue..

It seems to be Reach or upstream rather than a Telstraclear transparent proxy issue.

Lucky no one is holding their breath.

1 | ... | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12