Latency vs Ping

I have a reasonable ADSL connection (with EE/Orange) that gives 16Mbps down and 1 up. My question concerns the latency figure reported by the speed tests.

The ping time (using ping in a DOS cmd window) to a few servers such as BBC, google, thinkbroadband.com is consistently around 10-12 ms. However, the ThinkBroadband speed test gives a latency of 40-70 ms. The BT Wholesale and Which speed tests give a similar latency, but the Ookla tester gives 10 ms ping.

Are these testers measuring different things? Why is the ThinkBroadband latency so much more than the ping time?

These servers are all in different places. The further they are away from you, the higher the ping.

Whilst that could be true (e.g. if there were different numbers of routers between me and the servers), that is not the concern here. ALL of those example servers give a ping of the order of 10 ms, which I take as indicating that my ISP can get a packet from me to a website and back pretty quickly.

The question is why do the speed tests report a much higher latency than the ping time?

For example I have just pinged broadbandtest.which.co.uk in 10 ms and the speedtest run on my PC to that website reports latency 99 ms. A ping to thinkbroadband.com is 10 ms (min 9, max 13) but the speedtest has given a latency of 113 ms.

WIth out know exactly what the speed testers are measuring and reporting a latency it's hard to compare their figures against a ping.

A ping is a low level IP test and quite simple. A latency report might include extra things like setting up a TCP connection, or not!

In reply to a post by Thinker27:

A ping to thinkbroadband.com is 10 ms (min 9, max 13) but the speedtest has given a latency of 113 ms.

AFAIK the TBB speedtester has always given incorrect (higher) latency times, so I would take its ping results with a large pinch of salt. Why it churns out bloated ping times I've no idea...

Best way to measure your latency is by pinging a web/IP address preferably on a wired connection. I've found speedtest.net also gives you a reasonably accurate latency reading most of the time.

Our older flash speed test would give lower latency as you had better control of packet sizes, currently tester is a small HTTP TCP packet so a lot bigger than say an ICMP or UDP packet, hence why it is called Latency and never ping

The ookla figure may be their flash or app based tester

Because a webpage cannot issue ICMP pings

Agreed, I get about 22ms latency on the TBB Speed Tests (in my sig) and the following when I Ping the same server:

ping speedtest7.thinkbroadband.com

Pinging speedtest7.thinkbroadband.com [80.249.106.133] with 32 bytes of data:
Reply from 80.249.106.133: bytes=32 time=4ms TTL=53
Reply from 80.249.106.133: bytes=32 time=4ms TTL=53
Reply from 80.249.106.133: bytes=32 time=4ms TTL=53
Reply from 80.249.106.133: bytes=32 time=4ms TTL=53

Ping statistics for 80.249.106.133:
    Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
    Minimum = 4ms, Maximum = 4ms, Average = 4ms

I don't normally pay attention to the latency due to I know its normally in the 3 to 7ms.

Paul

As bsdnazz suggested, it would help to know exactly what the speed tester is testing regarding latency. I presume it is somehow measuring a transit time whilst the upload or download is in progress. This could be more "real world" in a sense, and presumably indicates a slowing down of the routers when under load.

Maybe it is useful for diagnostics, comparing different situations. However, I wonder how relevant it is as a measure of the quality of a broadband connection, in the sense of how well it performs as perceived by a user.

When streaming, the latency doesn't seem to be as important as speed, it just delays the time at which the download finishes by ~a tenth of a second. These days when loading a web page or transacting business, connections are made to lots of servers. I wonder whether ping times are more relevant to performance.

You are misunderstanding/overthinking it.

As MrSaffron has said, the speedtest doesn't use ICMP to measure latency. Doing a ping in command prompt is the most accurate method of determining the actual path of your connection to the thinkbroadband servers - and even that is only relevant as far as from you to TBB (however it does of course give you an indication of what your baseline is going to be).

In reply to a post by Pipexer:

You are misunderstanding/overthinking it.

As MrSaffron has said, the speedtest doesn't use ICMP to measure latency. Doing a ping in command prompt is the most accurate method of determining the actual path of your connection to the thinkbroadband servers - and even that is only relevant as far as from you to TBB (however it does of course give you an indication of what your baseline is going to be).

ping only uses a probable path that tcp might use,
and just as accurate is the tbb bqm which does ping from the far end.

Edited by ggremlin (Sun 14-Jan-18 22:11:42)

Latency of HTTP packets is actually pretty important, since web pages do lots of grabbing of smaller elements so latency as measured in the speed test is relevant, remember a webpage is just one URL but may contain 100 elements that need fetching.

The single figure reported is measured when the test was otherwise idle, there are additional measurements done during the download and upload phase to assess buffer bloat. The analysis button in the test shows the actual measurements in a graph.

I think the answer to the question must be that ping and latency are measuring different things. In terms of telephone technology a ping is the time it takes for the called phone to start ringing, not the time taken to answer it so that's pretty meaningless unless it's obviously very high. Perhaps someone knows what the TBB latency time actually measures?

What users are interested in is how long it takes to see a web page on screen. As the browser has to assemble most pages from numerous elements usually fetched from different servers that can be quite long. Some of those elements will be cached on intermediate servers along the path and others, such as your bank account, will not. Almost all elements will require a DNS lookup before they can be accessed so it's important to use a fast DNS server - again the ping time is not necessarily an indicator of time taken to provide data, though a short ping time will help.

It's useful to use traceroute to find how many hops to various destinations as a large number could indicate problems (congestion?) with your ISP's servers or others along the route. Wildly varying ping times are also an indication of problems.

>Perhaps someone knows what the TBB latency time actually measures?

How often do I need to repeat myself - round trip time for a small HTTP TCP packet.

Many thanks to all for your thoughts.

I think I was trying to be over-analytical.

Internet connections and usage are varied and complex. Like other measurements, perhaps ping and latency are best used subjectively, to make general comparisons over time and to give a pointer to a problem if things deteriorate.

In reply to a post by Thinker27:

Many thanks to all for your thoughts.

I think I was trying to be over-analytical.

Internet connections and usage are varied and complex. Like other measurements, perhaps ping and latency are best used subjectively, to make general comparisons over time and to give a pointer to a problem if things deteriorate.

yea, you need to consider transmission/re-transmission errors etc., as well as the load/performance of the remote servers
ping/latency are just two of the tools in the box