Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[craigski]

Be interesting. 660,000 packets a second being acknowledged by 34,000.

400:1 explains how Virgin Media sold 53 download on 1.75 upload and it worked fine 

Yes, may have worked a while back, before the proliferation of multi device households.

Yes interesting, but I don't think typical these days. The typical household would have multiple users using multiple devices, communicating to multiple services on multiple networks, all with packets of different sizes.

Just for kicks I looked at my router interface (uptime 25 days), I see following ratios:

packets Download:1 Upload:1.3
bytes  Download:1 Upload:2.8

[XGS_Is_On]

Was more a reference to the concerns some had about 1.75 not being enough to download at 53 due to acknowledgements overwhelming the upload than a reference to individual usage patterns.

[Chrysalis]

On gig1, the ack's are nowhere near close to saturating the upstream, I think I remember you telling me a while back VM do this magic and it seems to work reasonably well.

At 963mbps down in dumeter, it reported 8mbps up of ack bandwidth. Approx 120:1 ratio.  But thats just to my hub right, I think you explained the magic is done 'after' I see it, so will be less than actual 8mbit of US used?

[XGS_Is_On]

Yes. Standard for DoCSIS kit to act as a layer 4 proxy. Ack suppression.

[Weaver]

Agreed. Assuming one TCP ACK PDU for two DATA PDUs typical. Assuming the worst case inefficiency, IPv6+TCP with timestamps would be a 72 byte IP PDU containing a TCP ACK vs 1500 byte data IP PDU say. I make that 2.4% IP ACK PDUs / IP DATA PDUs byte count ie bandwidth. So bandwidth 1/x = 20.83 approx. Now SACK could of course increase that latter figure, as would IPv4 or no timestamps.

If they are doing L4 proxying then that’s potentially very naughty as the L4 reliability guarantee is broken. Unless of course the proxy is like a ‘custodian’ in the sense of the term in the bundle protocol, and in that case there needs to be a chained guarantee of successful forwarding. Not impressed by that at all.

[XGS_Is_On]

The cable modem ack suppression just accumulates acknowledgments. As it's a TDMA upstream it'll have a bunch of acknowledgements in its buffer that are unnecessary to send. It accumulates them and acknowledges all received data in one message. It forwards selective acknowledgements as necessary.

Acknowledging 2:1 throughout is entirely unnecessary. Only need to acknowledge with enough frequency to ensure windows don't fill. Other side is only concerned by the quantity of unacknowledged data, not by how many PDUs it came in. 2:1 only really useful at the start if slow start is being used to ensure windows grow.

[Weaver]

Agrees with XGS, I didn’t put it very well at all. What I was trying to say, and failed miserably, is that generating fake ACKs is naughty. Delaying them so that you can ACK for more recvd data is entirely sensible, as XGS says, as long as you don’t leave it way too late and blow it.

[XGS_Is_On]

CityFibre are now blowing fibre into their ducts serving me. Might well replace my Openreach backup with CityFibre. Not least because it means a diverse physical path following the CityFibre ducting rather than the Openreach ducts once out of the, small, development.

[Dwight]

Asymmetric DS/US ? 

It's still  awesome!

[XGS_Is_On]

I'm sure some would be offended that it doesn't even average 0.1% utilisation but it's good to have the capacity there on the rare occasion I need it for sure.

[Chrysalis]

CityFibre are now blowing fibre into their ducts serving me. Might well replace my Openreach backup with CityFibre. Not least because it means a diverse physical path following the CityFibre ducting rather than the Openreach ducts once out of the, small, development.

Will be interesting to see if you get activated earlier than my street, my street was done months ago.  Please keep us updated.

[XGS_Is_On]

On Tuesday they were enjoying the weather building the contents of the primary nodal cabinet: the one across the road. The secondary that serves us is already built.

I presume next will be a splicing crew, then the blowing of fibre to light CBTs.

[XGS_Is_On]

Because I'm an idiot and I've some 25G ports that need terminating I've purchased a QNAP QSW-M5216-1T. It is disappointing. It's had zero software updates since release and its support and status information for LAGs is unacceptable given the mass of information and options on the NAS they make. Indeed it could actually hamstring their NAS.

They are almost certainly using a Marvell Prestera switch chip so could be doing it better with hardware acceleration.

For now it's acceptable though this won't last much longer as the WAN breaches the 10G mark.

Wouldn't recommend that switch. A great bit of hardware with awful firmware support.

[burakkucat]

Because I'm an idiot and I've some 25G ports that need terminating I've purchased a QNAP QSW-M5216-1T.

. . .

Wouldn't recommend that switch. A great bit of hardware with awful firmware support.

Thank you for the latest update and the advice as what to avoid.

[XGS_Is_On]

Will be interesting to see if you get activated earlier than my street, my street was done months ago.  Please keep us updated.