BQM analysis

Check this snapshot graph:

My Broadband Ping

Nothing changes whatsoever during the working day other than it is used far less than in the evenings. The peak (if you can call such a straight increase a peak) appears to follow the capacity meter for my Entanet node. I think it's congestion. Is that more likely to be Entanet or the exchange? Packet loss from a router with nothing else to do apart from being pinged.

All ideas welcome.
.

I'm sure you're aware that the packet loss can occur anywhere between the tbb server and your router, and in either direction!

The tbb "reverse" traceroute tool might be useful in trying to work out where it's happening.

I'm sure you're aware that the packet loss can occur anywhere between the tbb server and your router, and in either direction!

Stares blankly into space......

To be fair, I've never quite got the jist of what it is trying to tell me, averages/best/worst etc.

How many cycles to extract the average?
.

If it is indeed correlated closely with the Entanet node loading, then you have answered your own question.

Remember the days of BT Centrals? An Entanet node is a very similar thing, but less easy to manage.

It is in fact one of 20 BT Wholesale nodes, and at each node Entanet have an MSIL, which is the 21CN equivalent of a Central. On that they rent a certain maximum throughput level, but whereas on a Central the absolute limit was 622Mbps, an MSIL is 1 (or I expect now available 10) Gbps. As there are 20 nodes, that is a lot of potential throughput, but unfortunately your connection is hard-wired to the MSIL.

If the load on the node exceeds the rented amount the BTW allow it to go through, as long as it is below the maximum possible, but IIRC above 105% of rented the cost goes up dramatically.

Quite how Enta are handling this problem these days I don't know, as I understand the MSIL "ALT" was discontinued long ago. Perhaps BTW have introduced a "hard" cap to solve it, or perhaps your node is at the full 1Gpbs rental and overloaded.

But it certainly sounds as though that is where the congestion is, at the Entanet MSIL.

(That sounds a bit garbled to me, as dealing with uncertainties, but I hope it is understandable).

In reply to a post by mrnelster:

To be fair, I've never quite got the jist of what it is trying to tell me, averages/best/worst etc.

How many cycles to extract the average?

From the setup page:

The Firebrick device sends out an ICMP echo request (a 'ping') packet and measures how long it takes for your router (or computer if you have a DSL modem) to respond. We then plot the graph based on 100 seconds' worth of pings for each point to show the minimum, maximum and the average (mean) latency. If any packets are dropped (i.e. we don't receive a response), this will be drawn in red from the top of the graph down.

The reverse traceroute tool can help to determine which part of the route is introducing delay/losing packets, but it needs running at frequent intervals to build up a picture and can be no more than a guide. The only way to be sure is to get the owner of the "slow" server/router to admit to it, and that can be tricky

Edited by billford (Mon 23-May-11 22:35:15)