ISP stats: Ping Latency and Packet loss.

[Fenix2]

Over the past few months I've been playing with a tool called Smokeping SmokePing - About SmokePing which generates nice pretty graphs of latency and packet loss.

I would like to share some of the graphs and comparisions for ISPs or unusual stuff you I see.

Please keep your throughput tests and those totally useless speedtests on other threads.

The first 3 charts are for the latency over 100Mbit to a Linksys Router with the WRT firmware. This is just pinging the local router over the local LAN using a Cisco switch. This customer has a 4/4M VDSL and is getting ripped off HK$7000 by PCCW. PCCW will solve the problem if you pay more for a better router..

These stats were collected on the 19th of April.

[Fenix2]

In those charts the latency increases to 16ms when someone is downloading heavily, otherwise it is 2ms!!!! If there are multiple users browsing the cpu on the router will not be able to keep up and max out at 200KB/s where it starts to lose data as you will see in a second. The circuit should be able to do 520KB/s.

Following two charts pinging much further outside the network to an ISP in London - LINX.

One is using the PCCW line when someone is doing the 200KB/s a single file over HTTP, with one socket. The other is Diyixian. Both circuits are 4/4M. I have 2 Smokeping sessions which send 20 pings ever 5 minutes.

As you can see the PCCW has mass loss, the issue is that the latency and packet loss push the overall latency above 500ms. It seems that when latency goes over 500ms = guaranteed packet loss.

The instability of the circuit is not PCCWs back bone issue it is the crap equipment provided by PCCW. However it is clear that Diyixian is 70ms faster to this destination, which wasn't the case before. PCCW changed its backbone to higher latency links and probably cheaper ones last year.

I hope to have another chart for HGC to compare with in a couple of weeks.

[MrMoo]

The first 3 charts are for the latency over 100Mbit to a Linksys Router with the WRT firmware. This is just pinging the local router over the local LAN using a Cisco switch. This customer has a 4/4M VDSL and is getting ripped off HK$7000 by PCCW. PCCW will solve the problem if you pay more for a better router..

Is there a point in this post somewhere?

[Fenix2]

The point is maybe we share some actual figures and charts instead of hearsay information and opinions.

Got any charts on HGC?

[intsangity]

Got any charts on HGC?

What about Wharf? Are we just interested in business DSL or residential DSL as well? How about the DIA packages where they run fiber to the office? I recently got quotes for 10Mbps DIA circuits for around 7k/month.

I can't believe that people are paying 7k for 4/4 VDSL. I wish I was making that spread.

[Fenix2]

Wharf? well that is the whole point of this thread to share information. You can all install this package and see what numbers you get.

BTW, I don't have access to any wharf circuits, maybe someone else can.

Maybe we should have some standard set of IPs which we conduct the tests against so the reports are all valid.

I think I should set things up to start publishing stats for my beloved Netfront, their problems are laughable right now.

[jgl]

I had trouble following the first two posts. Is the point of it that a Linksys router can't keep up with high throughput, or is it something to do with the actual links themselves?

[shri]

No Windows version? Just checking to see if I've missed something. Will install later on a linuxy system and give it a shot...

[Fenix2]

Shri here are instructions how to do it in Windows.

Ogi How to install smokeping under windows

Projects Log Blog Archives How to install smokeping under windows

[Aidan]

I'd like to see this run on a few different regular home plan DSL / Wireless connections at different speeds to see the latency (especially to see if it changes during BT / other kinds of downloads as evidence of traffic shaping / throttling). I have seen letancy go through the roof during some downloads (far more than could be accounted for by serialisation delay / congestion as the bw was a tiny fraction of the connection speed and it was the latency to the first edge router that shot through the roof)