Print

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

[Weaver]

I'm going mad. Bonkers.

Today I've tried half a dozen different (non-flash, ipad) speed tester websites and apps. All but two of them report normal downstream numbers.

Two have suddenly gone very weird. I'm using Safari on an iPad.

The first is :
    http://sod.ms/fast
The second is :
    http://speedof.me/m/
(the latter part /m/ I suspect is for supposed mobile, mis-sniffed iPad)

These two report speed figures _down to 30-50% of normal_ downstream. (E.g. 1.4Mb/s instead of 5.3.)

And this is just suddenly different today.

What on earth might be going on?

[Weaver]

Unfortunately its still happening today. Any ideas?

[Weaver]

NOW DISAPPEARED!!

The issue has now vanished mysteriously. I'm inclined to blame either the ISP or some common point further out in the internet, can't say which, else how could two totally different services go bad simultaneously?

[Weaver]

Update. I did a traceroute to sod.ms, this doesn't leave the ISP’s own network, its an Andrews & Arnold-internal website (their little rather juvenile joke in the name), so it has to be A&A's fault, in that a route that affects both sites, one external and one internal, had something bad happening.

[kitz]

I saw your post the other day, but wasnt able to do any testing because I needed to be on the ipad and in the same room as the router to ensure that it wasnt anything to do with the wifi.

From ipad

sod.ms gives me 44.36Mbps down and 526 Mbps up
speedof.me  -  I tried that a couple of times and the test wont complete "A problem occurred with this webpage".
Ookla mobile speedtest -  37Mbps down 18.7Mbps up

From desktop

sod.ms -  55.097677Mb/s 55097677 468660 4692440 0.630081
speedof.me - crashed firefox first time.   Tried again got 78.9/16.83 Mbps.   The 78.9 is a bit generous.  Usual speed is circa 73.5 and even allowing for TCP overheads I'd expect a max of 78Mbps.
ookla speedtest.net - 73.24/18.98 Mbps

[Weaver]

Of course these testers might well be using completely different methodologies, measuring different things, servers in unfavourable locations. But the numbers you are seeing are consistently low or high relative to one another, sod.ms is always low, speedof.me is always high and unbelievable. I also wonder if some are applying multiplication fudge factors to derive say sync rate from ip payload or tcp payload.

These testers are pretty useless imo unless you know how to apply fudge factors to normalise them with a vslue you have obtained by other means, such as meditation.

[burakkucat]

I trust none for an absolute value but will continue to use my favourites and will trust the relative results.

[NewtronStar]

Eye ye canny beat the BT speed tester it's mustard 

[Weaver]

Update.

The other day a modem “went bad” on me in that it started racking up a lot of ds errored seconds after being turned off for quite a while then back on. There was accompanying bad ds packet loss.

    http://forum.kitz.co.uk/index.php/topic,15811.msg294533.html#msg294533

Now I got an error message from amazon streaming video warning me about a lack of bandwidth. This was the first thing that alerted me to the problem. So I tried the speedof.me speed tester  which read ds 0.8 Mbps instead of the usual 5.2. Andrews and Arnold’s line quality monitoring had lit up red showing loss of PPP LCP “pings”.

So it seems that the speedof.me speed tester reacts this way to heavy packet loss. Perhaps this was the explanation for the original episode.

[NewtronStar]

All I know is errored seconds and would not know what a bad ds packet loss looked like even if it hit me in the face, what does it feel like and how does it effect your broadband ?

[Weaver]

@newtronstar

The downstream packet loss is a possible cause of breakup of streaming video or inbound voice over ip. Congestion is an alternative possibility.

On the Andrews and Arnold continuous line monitoring graphs

    http://support.aa.net.uk/CQM_Graphs

you can see this kind of problem showing up as "dripping blood" bright red coming streaks coming from the top of the graphs indicating loss of ppp lcp probe responses. (These ppp probes are analogous to pings done with icmp.)

[kitz]

@NS

TBB also do a free graph for latency monitoring and packet loss.    Downside is that you need a static or sticky IP.  Its no use if your IP changes every time you resync.

[Weaver]

re: Kitz' tip about the thinkbroadband quality monitoring service

Excuse me for wandering off topic for a small fyi

I seem to remember this service is provided by a Firebrick 6000-series LCP “pingbox” which is basically a server that can probe umpteen lines per second, designed by the Firebrick partnership who are Watchfront plus Andrews & Arnold doing all the hardware and software from scratch.

So, yay ! for a British-designed hardware product brought to you by two of your favourite ISPs who use their own kit to run their own operations as well as selling it to the market.

Some other ISPs use Firebrick 6000-series _routers_ (and various other kinds of boxes as well), I forget who.

[I use a Firebrick small router+firewall (plus lots of other s/w goodness), which I bought straight from A&A. It’s a good feeling being able to talk to the people who designed your router, and who are also your ISP, as you know where the buck stops if there should be any weirdness.]