Which speed test to believe? OOKLA or nPerf?

Forums › New Zealand Broadband › Which speed test to believe? OOKLA or nPerf?

Sideface

9649 posts

Uber Geek
+1 received by user: 15595

Trusted

Lifetime subscriber

#191142 25-Jan-2016 15:36

Like many other Geekzoners, I do a lot of speed testing.

I'm on "100/10 Mbps" HFC cable, which has had a lot of problems over the last 12 months (see Krapi Cable Congestion).

I often use the OOKLA speedtest (which uses Flash) as well as the nPerf speedtest (no Flash required). I'm also a TrueNet volunteer, so i have a third source of speed tests.

I find that OOKLA often gives much "better" (or more optimistic?) speeds than nPerf. eg downstream with OOKLA 101 Mbps vs nPerf 80 Mbps or less (both to fast Auckland servers).

In my experience, nPerf is closer to real-world performance (eg downloading a MS ISO) than OOKLA.

What is the experience of other Geekzoners?

Sideface

PeterReader

6028 posts

Uber Geek
+1 received by user: 461

Trusted

Geekzone

Lifetime subscriber

#1478331 25-Jan-2016 15:36

Hello... Our robot found some keywords in your post, so here is an automated reply with some important things to note regarding broadband speeds.

If you are posting regarding DSL speeds please check that

you have reset your modem and router
your PC (or other PCs in your LAN) is not downloading large files when you are testing - you are not being throttled by your ISP due to going over the monthly cap
your tests are always done on an ethernet connection to the router - do not use wireless for testing
you read this topic and follow the instructions there.

Make sure you provide information for other users to help you. If you have not already done it, please EDIT your post and add this now:

Your ISP and plan
Type of connection (ADSL, ADSL2, VDSL)
Your modem DSL stats (do not worry about posting Speedtest, we need sync rate, attenuation and noise margin)
Your general location (or street)
If you are rural or urban
If you know your connection is to an exchange, cabinet or conklin
If your connection is to a ULL or wholesale service
If you have done an isolation test as per the link above

Most of the problems with speed are likely to be related to internal wiring issues. Read this discussion to find out more about this. Your ISP is not intentionally slowing you down today (unless you are on a managed plan). Also if this is the school holidays it's likely you will notice slower than usual speed due to more users online.

A master splitter is required for VDSL2 and in most cases will improve speeds on DSL connections. Regular disconnections can be a monitored alarm or a set top box trying to connect. If there's an alarm connected to your line even if you don't have an alarm contract it may still try to connect so it's worth checking.

I recommend you read these two blog posts:

I am the Geekzone Robot and I am here to help. I am from the Internet. I do not interact. Do not expect other replies from me.

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

Behodar

11089 posts

Uber Geek
+1 received by user: 6069

Trusted

Lifetime subscriber

#1478362 25-Jan-2016 16:10

Ookla tends to measure the "last mile" of your connection due to its large network of servers. I'm not familiar with nPerf, but a service with fewer servers will probably be a better representation of the real world.

darylblake

1172 posts

Uber Geek
+1 received by user: 410

Trusted

#1478438 25-Jan-2016 17:01

Personally I think that cable network might actually get better. With Vodafone planning some upgrades and a lot of people who can get UFB opting to leave cable to take up UFB.

But yeh your speed test is as good as the last mile. Its a complicated kettle of fish, you might only get 10mbit to parts of the usa or europe, it depends who the ISP peers with, what paths the traffic takes and how much transit they have and their peering partners have at certain points across the web.

You should get a speed close to what you are paying for to servers in NZ. And the difference between ookla and nperf will be so small.

Talkiet

4819 posts

Uber Geek
+1 received by user: 3934

Trusted

#1478440 25-Jan-2016 17:03

Both give you a calculated estimate of maximum possible access speed - nothing more. Ookla in particular (speedtest.net) uses a non centrally tended subset of measures and then multiplies the result by a fudge factor before showing you the number.

Speedtest.net servers are also massively variable between sites. I have to say the Spark speedtest servers are probably the best in NZ (I can see the load on them and I can see the results for all tests run against them and even the Akl server shows a LOT of Gigatown users record speeds over 920Mbps).

I like nperf too, but all speedtests are optimistic and virtually never represent likely real world results, for one reason or another. (I include truenet results in this as well)

(I'm not saying any of the above aren't useful, but you can't compare them to each other because they measure different things, and you can't use them to estimate real world speeds)

Cheers - N

Please note all comments are from my own brain and don't necessarily represent the position or opinions of my employer, previous employers, colleagues, friends or pets.

Sideface

9649 posts

Uber Geek
+1 received by user: 15595

Trusted

Lifetime subscriber

#1483448 2-Feb-2016 12:18

Talkiet: <snip> Speedtest.net servers are also massively variable between sites. I have to say the Spark speedtest servers are probably the best in NZ ...

I'm in Wellington on VF HFC cable (which suffers from evening congestion), and regularly test:

1. OOKLA speedtest to VF Wellington server - this tends to give the "fastest" result

2. OOKLA speedtest to Spark Auckland server - typically slightly slower than the VF Wellington server

3. nPerf to MyRepublic server in Auckland (the only NZ server used by nPerf) - slower than both of the OOKLA speedtest servers

Sideface