*** Official Hyperoptic Discussion Thread ***

Associate
Joined
4 Jul 2008
Posts
1,997
Location
London
Hey guys,

I have hyperoptic in my building, and I'm currently using their router: Tilgin hg2381

It's been ok so far, but I noticed when I was uploading files to dropbox, everything else stopped working for other devices.

I'm thinking of upgrading the router to Google Wifi, does anyone have any experience with this, and it is simply just plugging in the new router, and you're all set to go?

Thanks!
 
Soldato
Joined
8 Nov 2006
Posts
22,966
Location
London
So turns out my cheap TP-Link router refused to sync above 100Mbps even on the 500Mbps package.

So I looked for an alternative.

I saw the old Google Router and the new Google Nest Router, but thought I'd try out the new Amazon Eero.

https://www.amazon.co.uk/eero-mesh-...ef=sr_1_6?keywords=eero&qid=1577465559&sr=8-6

Syncs at 500Mbps and the specs are good enough to get that across wifi. £50 saved.

Also a smart router controlled via an app, and you can create a mesh if you wanted.

For a 1Gbps connection you would need the Eero Pro which is actually more expensive than the Google Nest Router.
 
Soldato
Joined
23 Jun 2005
Posts
5,454
Hi all,

I'm moving into a new build in summer, and Hyperoptic is in place for all properties on the estate.

I currently use PlusNet coupled to an TP-Link VR900 router with no issues. Can I use my VR900 with Hyperoptic, and how does the service compare to plus net?
 
Soldato
Joined
15 Dec 2008
Posts
2,681
Location
London
I can't see why not - the HyperOptic service (well, the one I have) is just an Ethernet termination, so straight in to the back of that router and off you go.

I've no real experience of Plus NET but I've been with HyperOption for 3-4 years now, and they've been great. I've had to call them a couple of times (CGNAT was causing me some pain), and the people on the front line were quick enough to pass me on to people who knew their stuff rather than running through a script.

Big thumbs up from me tbh.
 
Associate
Joined
28 May 2007
Posts
634
Hi all,

I'm moving into a new build in summer, and Hyperoptic is in place for all properties on the estate.

I currently use PlusNet coupled to an TP-Link VR900 router with no issues. Can I use my VR900 with Hyperoptic, and how does the service compare to plus net?

Compared to plusnet, no contest. Lower pings, and regularly get over the 150meg up and down that my current connection speed gets me. When I had the gigabit speed option it was insane how rapid it was --ultimately I didn't need the extra speed so am spending less on the cheaper package.

Any router that's got an ethernet WAN port can be used, although many cheaper routers might start to choke on faster than 200-300mbit; since FTTC connections are 80/20 at most then this won't be noticed readily, but suddenly throw a 500/500 or 1000/1000 connection in the mix and you'll quickly be able to tell if your router hasn't enough grunt to keep up.
 
Soldato
Joined
23 Jun 2005
Posts
5,454
I can't see why not - the HyperOptic service (well, the one I have) is just an Ethernet termination, so straight in to the back of that router and off you go.

I've no real experience of Plus NET but I've been with HyperOption for 3-4 years now, and they've been great. I've had to call them a couple of times (CGNAT was causing me some pain), and the people on the front line were quick enough to pass me on to people who knew their stuff rather than running through a script.

Big thumbs up from me tbh.

Compared to plusnet, no contest. Lower pings, and regularly get over the 150meg up and down that my current connection speed gets me. When I had the gigabit speed option it was insane how rapid it was --ultimately I didn't need the extra speed so am spending less on the cheaper package.

Any router that's got an ethernet WAN port can be used, although many cheaper routers might start to choke on faster than 200-300mbit; since FTTC connections are 80/20 at most then this won't be noticed readily, but suddenly throw a 500/500 or 1000/1000 connection in the mix and you'll quickly be able to tell if your router hasn't enough grunt to keep up.

Thank you - the VR900 is a pretty good router so it should be up to the job. I will be opting for the 150mb package as there's only two of us in the household and my current 60mb speed is enough for our needs.
 

~cw

~cw

Associate
Joined
2 Jan 2019
Posts
164
Hi all,

I'm moving into a new build in summer, and Hyperoptic is in place for all properties on the estate.

I currently use PlusNet coupled to an TP-Link VR900 router with no issues. Can I use my VR900 with Hyperoptic, and how does the service compare to plus net?

Yes, you can! The VR900 LAN4 doubles as a WAN port. Reconfigure it as a cable modem router through the wizard, set it to "wireless router mode".

Be mindful that if you have speed or connectivity issues, Hyperoptic won't support you unless you have the Hyperoptic router connected to the Hyperoptic ethernet socket. Honestly, their new router (a ZTE H298A) isn't that bad... Way better than the Tilgin HG2381 which kept on locking up on me. I went through two before they sent me a ZTE :D

You could always daisychain the Archer off the HO router (disable the Archer's DHCP server / set it to forward DHCP traffic to the Hyperoptic router) and run a long cable between them. Then you effectively have a gigabit network switch inline but separate from the Archer kit which can be useful for diagnostics (or for more capacity).

Archer manual has the instructions for configuring as router - https://images-eu.ssl-images-amazon.com/images/I/D1ECr5jOjGS.pdf (page 19 of the PDF, "page 15" going by the footer numbers).

Looks like it's a dual stack device as well! Hyperoptic uses SLAAC for IPv6 so you should get DHCP-assigned IPv6 and IPv4 automagically.

Note that unless you pay £5 a month (yes, ripoff, I know) your IPv4 address will be via Carrier Grate NAT (CG-NAT), so you can't connect in from the outside world. Hyperoptic aren't unique in doing this but it is noteworthy.

The IPv6 address is internet-facing but you'll need a 6-to-4 gateway if you want to rely on that for inbound port forwards or hosting anything accessible from the web.


Speed and service-wise, Hyperoptic are waaaaaaaaay better than any VDSL/ADSL provider. It's fibre to the premises, the termination and building distribution will be slightly different depending on where you live. For me, it's literally a fibre pull straight to my property then media converted to ethernet through a Huawei switch in a small cabinet. You will quickly realise just how bad Windows is at sustaining high bandwidth throughput. If you do any serious network speed tests (e.g. with iperf), use a Mac or Linux machine.



Make sure you get referred by one of us (happy to refer you myself) as both people get a referral bonus. Also, if you're not sure about which package is best for you, go for a higher tier package on a one month contract then downgrade and sign up for 12 months if you're not sure.

I have the gig package on a deal and frankly it's worth it. Just for the way fast upload, it makes device backups, remote access (SSH/VNC/FTP) and heavy bandwidth usage so effortless. In fairness the 150 meg packages are also very zippy, but if you can get gigabit... Why not?

Their phone service is an amusement if you're interested, +£1 a month and they provide it via SIP (only accessible 'on-net' so you can't use it elsewhere unless you VPN back in to your home network).

FEB20 offer code deals are on at the moment as well. https://hyperoptic.com/offer/

(PM me if you want a referral!)
 
Associate
Joined
30 Nov 2005
Posts
507
Location
London
Just had Hyperoptical Activated after moving into a new flat. Initially the connection was quite erratic but after a conversation with a nice chap on the phone Sunday evening seems to have it consistently hitting 150/160mbs.

One issue which I cannot get to the bottom of however is whilst downloading games from Steam or from Microsoft on the xbox.

The download starts fine, I presume due to all the buffered data. But after a little while, generally a few minutes the speed drops off drastically.
Even down to KBs then slowly builds up again but then drops off again. The huge drops are not always sequentially, sometimes there is a huge drop in speed.
On steam I have tried different regions to download the files. I have reduced the firewall level on the router and placed it back to no effect. Even turned off or restricted
the windows defender firewall. Nothing seems to help. One thing that is really weird is when I use UBISOFTS game launcher to download game titles it is a solid as a rock.
Maximising the full bandwidth. I have tested UBISOFT and Steam and the xbox at varying times of the day and night (I do game quite late some nights 2/3/4am.) With the same results.

I have just switched out the CAT5 cables to a higher quality and speed CAT7 with the same results. I know it is overkill but though it might have been something to do with my cable quality.

Does UBISOFT use a different method to downloading games than Steam or Microsoft? If there is anything anybody could do or suggest I try and I would be so happy as this is the only
thing which is proving to be a problem with the connection now.
 
Soldato
Joined
9 Jun 2011
Posts
3,598
Ah wish i had hyper optic. BT are milking us for 330/50 soon to be 1gbps/150mbs.... Bring on Citybre of Hyperoptic, Ashame their model requires a block of flats/business!
 

~cw

~cw

Associate
Joined
2 Jan 2019
Posts
164
Just had Hyperoptical Activated after moving into a new flat. Initially the connection was quite erratic but after a conversation with a nice chap on the phone Sunday evening seems to have it consistently hitting 150/160mbs.

One issue which I cannot get to the bottom of however is whilst downloading games from Steam or from Microsoft on the xbox.

The download starts fine, I presume due to all the buffered data. But after a little while, generally a few minutes the speed drops off drastically.
Even down to KBs then slowly builds up again but then drops off again. The huge drops are not always sequentially, sometimes there is a huge drop in speed.
On steam I have tried different regions to download the files. I have reduced the firewall level on the router and placed it back to no effect. Even turned off or restricted
the windows defender firewall. Nothing seems to help. One thing that is really weird is when I use UBISOFTS game launcher to download game titles it is a solid as a rock.
Maximising the full bandwidth. I have tested UBISOFT and Steam and the xbox at varying times of the day and night (I do game quite late some nights 2/3/4am.) With the same results.

I have just switched out the CAT5 cables to a higher quality and speed CAT7 with the same results. I know it is overkill but though it might have been something to do with my cable quality.

Does UBISOFT use a different method to downloading games than Steam or Microsoft? If there is anything anybody could do or suggest I try and I would be so happy as this is the only
thing which is proving to be a problem with the connection now.

I've noticed Ubi downloader is sometimes slightly faster than Steam, possibly because it's even more aggressive than Steam in how it opens connections or transfers data. Late in the evening on the server closest to you, speeds from Steam can actually be as high / higher.

Run Task Manager and monitor your CPU while downloading. If it immediately spikes to around 100% while downloading, you're suffering from Windows-related problem(s).

What speed test results do you get over wired connection from http://speedtest.hyperoptic.com/ ? Run it half a dozen times and it'll test to their on-network server.

Install LatencyMon and run it while you're doing high speed downloads. On my machines, ndis.sys causes instant increases in DPC latency, indicating that ndis.sys is the culprit. You may find a similar driver is causing stuttering/high CPU usage, which in turn causes a loss of network performance.



A few other factors can influence speeds on ultrafast lines. What type of antivirus/firewall being used (have you tried with it fully disabled?), speed of machine, which OS you're using (Win10?), what kind of network adapter you're using (plus driver version) and even what kind of CPU you have.


Ultrafast connections like Hyperoptic, Cityfibre, Virgin Media etc are described as Long Fat Networks (LFNs) because they're actually quite lossy, involve many hops to reach the destination (including multiple peers to transit providers) and the networks are very contended. They're high bandwidth and high latency - quite variable overall performance. This makes it harder for Windows' network congestion control protocols.

Also, IMO Windows is REALLY non-optimised for high speed data throughput over LFNs. It's fine on a LAN with 2 ms latency, but as soon as you introduce more hops and latency -- bam! Welcome to the wonderful world of woeful Windows windowing.

Basically, it's the same problem I've noticed (as well as quite a few other people who have done high bandwidth utilisation tests) and investigated in some detail.


What I think might be happening to you may be the same thing which happens to my main Windows PC. It's incapable of sustaining a high throughput on a single thread, due to inadequacies with its TCP windowing and congestion control algorithms. On my main machine - a Win10 Pro running i7-6800K with an Intel NIC - I get such severe CPU bottlenecking over 600 mbit/sec that I'm frequently unable to get steady >750 Mbps continuous on a CDN download on a gigabit line. Same with upload. I've spent weeks investigating, testing different network adapter settings, driver versions, running iperfs to various high-bandwidth servers... Similar results each time. Doesn't matter about time of day or which test servers I use.

And on top of that, some versions of network adapter drivers introduce significant extra latency - the latest version of Intel drivers for my chipset cause MASSIVE lag spikes, DPC latency increase and audio stuttering at high speeds, including over the LAN. This happened with several different firewall/AV products and also with them disabled.


If you can temporarily boot to Linux (make a Ubuntu or Mint Linux LiveUSB with a persistent storage volume, they're genuinely useful!) run some speed tests. Use peedtest.net, http://speedtest.hyperoptic.com and also run some single threaded iperf tests to a fast iperf server where you can watch performance in real time. Compare the TCP retries and bandwidth utilisation.

Generally Linux can reach full speed almost instantly with minimal packet retries. In Windows, unless you have a unicorn system (perfect NIC drivers, no other overhead and ideal network conditions) you will see bandwidth rise then rapidly reduce, repeatedly, and CPU usage will be all over the place.

Also check https://store.steampowered.com/stats/content/ to see what times are quietest for your region...


The slow XBox downloads are probably just Microsoft's CDNs being too busy to handle traffic. :D

Which model Hyperhub do you have?
 
Associate
Joined
30 Nov 2005
Posts
507
Location
London
I've noticed Ubi downloader is sometimes slightly faster than Steam, possibly because it's even more aggressive than Steam in how it opens connections or transfers data. Late in the evening on the server closest to you, speeds from Steam can actually be as high / higher.

Run Task Manager and monitor your CPU while downloading. If it immediately spikes to around 100% while downloading, you're suffering from Windows-related problem(s).

What speed test results do you get over wired connection from http://speedtest.hyperoptic.com/ ? Run it half a dozen times and it'll test to their on-network server.

Install LatencyMon and run it while you're doing high speed downloads. On my machines, ndis.sys causes instant increases in DPC latency, indicating that ndis.sys is the culprit. You may find a similar driver is causing stuttering/high CPU usage, which in turn causes a loss of network performance...........

.......The slow XBox downloads are probably just Microsoft's CDNs being too busy to handle traffic. :D

Which model Hyperhub do you have?

Thanks for your detailed reply ~cw. I will source all the info tonight if possible, the girlfriend has me pencilled in for IKEA construction
but will try and get some time in to reply as best I can. Obviously can detail system spec and Router version when I get home.
Some of the other stuff went over my head. Off the top of my head the upload was around 180 and the download was quite low at 80-90mbps.
That prompted the call to their support on sunday. It seems the bloke sorted it out as it has been reasonably symmetrical. at around 160 up and down.

Will post back as soon as I can

cheers

B
 
Associate
Joined
30 Nov 2005
Posts
507
Location
London
Ok I have taken a few screenshots and tried to detail as much as I can.

Router Version: ZXHN H298A (software version V1.0.x)
Windows 10 Pro 64 Bit (10.0.18363 Build 18363)
Asus Maximus IV Extreme-Z Intel Z68 (Stock BIOS 210)
Intel Core i7 2600K 3.4 GHZ
Crucial RealSSD M4 256GB SATA (OS)
Crucial MX200 1TB SSD (Steam Folder)
WD 300 GB Velociraptor (storage)
WD 150 GB Raptor (storage)
EVGA GTX 980Ti Superclocked+ ACX 2.0+ 6144MB
Corsair AX1200 CMPSU
16GB (4x4GB)Corsair Dominator DDR3 CMP8GX3M2A1600C9



I have conducted 3 speedtests via Hyperoptic and also opened the task manager whilst downloading from Steam.

UK London Server around 20.10. So you can see the CPU utilisation.

My network adaptor is the onboard : https://www.asus.com/ROG-Republic-Of-Gamers/MAXIMUS_IV_EXTREMEZ/specifications/


Direct link to image Gallery. Hope some of these are of use.

https://postimg.cc/gallery/2am8aut60/

Cheers

B
 
Caporegime
Joined
18 Oct 2002
Posts
26,053
Plug your PC straight into the Hyperoptic wall socket, there's nothing clever about the routers. If there is a repeatable improvement in your download speeds then the issue is either your router or some other device connected to the router hogging the connection. If there's no improvement then the issue is with your PC or with the Hyperoptic service.
 

~cw

~cw

Associate
Joined
2 Jan 2019
Posts
164
Ok I have taken a few screenshots and tried to detail as much as I can.

Router Version: ZXHN H298A (software version V1.0.x)
Windows 10 Pro 64 Bit (10.0.18363 Build 18363)
Asus Maximus IV Extreme-Z Intel Z68 (Stock BIOS 210)
Intel Core i7 2600K 3.4 GHZ
Crucial RealSSD M4 256GB SATA (OS)
Crucial MX200 1TB SSD (Steam Folder)
WD 300 GB Velociraptor (storage)
WD 150 GB Raptor (storage)
EVGA GTX 980Ti Superclocked+ ACX 2.0+ 6144MB
Corsair AX1200 CMPSU
16GB (4x4GB)Corsair Dominator DDR3 CMP8GX3M2A1600C9



I have conducted 3 speedtests via Hyperoptic and also opened the task manager whilst downloading from Steam.

UK London Server around 20.10. So you can see the CPU utilisation.

My network adaptor is the onboard : https://www.asus.com/ROG-Republic-Of-Gamers/MAXIMUS_IV_EXTREMEZ/specifications/


Direct link to image Gallery. Hope some of these are of use.

https://postimg.cc/gallery/2am8aut60/

Cheers

B

How was the DIY? ;)

Are you on the 150 meg or 500 meg package? If you're on 150, that looks like you can max your connection OK with multi-threaded downloads... CPU usage looks fine for the Steam downloading, not bouncing off 100% which would indicate a significant OS/driver problem, so more investigation needed.


In the Steam screenshot, your download at the moment you took the screenshot is 15 MB/sec, about 120 Mbits/second. That's decent for a 150 Mbit/sec connection, if you consider PC overheads, TCP protocol overheads etc. I can't get a full gigabit/sec of data after overheads, even on a Linux machine which can easily do full gigabit/sec over LAN. However with iperf I can often max the line out at a stable speed between 109 and 115 MB/sec (depending on test server).

There's a lot involved with stable high-speed Internet transit. I've noticed that Hyperoptic's peering decisions can sometimes be a bit 'weird'. But I think the initially higher peak speed you see is possibly how the connection behaves, rather than a quirk of your setup. I've rarely been able to achieve anywhere near the full line rate of my connection with a single-threaded connection, except on an iperf to a server on a high quality 10gig/sec link.

I notice that on the speed.hyperoptic.com multi-threaded (Ookla-based) tester, the download and upload speed test do that nice curve almost before stabilising at a pretty solid speed. If you're on the 150 Mbit/sec package, that shows you're able to get maximum achievable throughput at least, plus a bit extra to accommodate TCP overhead (how generous of them).

The graphs look like you're getting an initial burst of higher throughput before it settles down. 167 Mbit/sec is actually about 20 MB/sec, 150 Mbit/sec is 18.75 MB/sec (including overheads). The brief peak of 24 MB/sec to me looks like a calculation error.

If you do a single-thread download in your browser from one of the following test servers while monitoring your bandwidth (I like NetMeter Evo), how does it graph?

http://ams-speedtest-1.tele2.net/1000GB.zip
http://speedtest-ams2.digitalocean.com/1gb.test
http://ipv4.download.thinkbroadband.com/1GB.zip

http://ipv6.download.thinkbroadband.com/1GB.zip
http://176.67.170.244/speedtest.256mb (Manchester), http://109.123.87.183/speedtest.256mb (London)

As Caged mentioned, hook straight in to your Hyperoptic socket and test also. If the Hyperhub is misbehaving, you've eliminated it. Which model Hyperhub do you have? The Tilgin or one of the two ZTEs?


I haven't tested on the slower Hyperoptic packages, always had the gigabit. However this might just be normal behaviour for your connection. I presume lower bandwidth packages are done by HO using policer queues (or traffic shaping) on the switch your socket connects to. If so, that initial spike of higher bandwidth (smoothly curving off to your average speed) may actually be normal and a characteristic of your traffic as the speed settles down to the committed rate. The traffic management policy may permit a brief spike of higher speed before the policies take effect.



A way to test your true burst and sustained bandwidth would be to do repeated iperf3 tests to and from a public iperf server. (see https://iperf.fr for publicly accessible iperf3 servers and instructions on use). If the speed tails off in the same way, it's probably just a characteristic of how Hyperoptic manages the line speed.

This is why I suggest making a Linux LiveUSB stick and booting from that for tests, it totally eliminates Windows from the equation. I found my NIC and overall network performance was way better under Linux. Better network stack, fewer overheads.


Other possibilities are that your building's incoming bandwidth is almost at capacity and you're being locally traffic managed for certain classes of traffic. I believe in some installations with high numbers of customers sometimes bandwidth suffers a bit at peak time, but then you'd probably notice that on all types of usage.

During a long period of downloading or uploading, with a speed graph app (like NetMeter Evo I mentioned above) watch the speeds. If you get sudden drops to almost zero then a recovery back up to a normal speed, that's a possible problem with your CPU setup or drivers/DPC/packet loss. Less likely is something physical like a dodgy patch to the Hyperoptic's upstream network switch. It may also potentially be contention on the incoming link supplying the whole building.

If you're running Windows 10, also install LatencyMon and run it while doing a heavy download. If you get significant DPC latency issues and warnings on screen, it's your PC unable to keep up for some reason, usually a driver-related issue.



Slightly related -
I did a bunch of testing to see what would minimise OS and driver impact on network throughput. On my network adapter, for better overall performance (this is what Intel and a few research organisations recommend for more reliable high-speed throughput), disabling the offloading features of your network adapter can reduce latency and improve performance (at a possible hit to CPU usage). This is in the NIC's Advanced tab...

Make a note of what settings you were using before changing so you can revert if no difference! This is how I've configured my NIC, an onboard Intel I218-V:
(settings not mentioned are inconsequential or left at defaults)

  • Adaptive Inter-Frame Spacing: enabled
  • Flow Control: Rx & Tx Enabled
  • Interrupt Moderation: Enabled
  • Interrupt Moderation Rate: Adaptive
  • IPv4 Checksum Offload: Disabled
  • Large Send Offload V2 (for IPv4) and (IPv6): Enabled (...though some people recommend Disabled, I found it slightly increased my CPU usage)
  • Receive Buffers: as high as possible (I can get to 2048, you may only be able to do 512 or 1024)
  • Transmit Buffers: as high as possible
  • TCP Checksum Offload: Disabled (for IPv4 and IPv6)
  • UDP Checksum Offload: Disabled (for IPv4 and IPv6)
I run the 29/09/2017 drivers (core driver version 12.17.8.7, PROWin v22.10) because later revisions appeared to be causing massive DPC latency problems. Stuttering, audio issues, momentary freezes etc.

I don't think changing these settings on your card will make much if any difference. But thought it worth making a note for others on the gigabit package who might be going through the same testing I did.
 
Associate
Joined
30 Nov 2005
Posts
507
Location
London
How was the DIY? ;)

Are you on the 150 meg or 500 meg package? If you're on 150, that looks like you can max your connection OK with multi-threaded downloads... CPU usage looks fine for the Steam downloading, not bouncing off 100% which would indicate a significant OS/driver problem, so more investigation needed.


In the Steam screenshot, your download at the moment you took the screenshot is 15 MB/sec, about 120 Mbits/second. That's decent for a 150 Mbit/sec connection, if you consider PC overheads, TCP protocol overheads etc. I can't get a full gigabit/sec of data after overheads, even on a Linux machine which can easily do full gigabit/sec over LAN. However with iperf I can often max the line out at a stable speed between 109 and 115 MB/sec (depending on test server).

There's a lot involved with stable high-speed Internet transit. I've noticed that Hyperoptic's peering decisions can sometimes be a bit 'weird'. But I think the initially higher peak speed you see is possibly how the connection behaves, rather than a quirk of your setup. I've rarely been able to achieve anywhere near the full line rate of my connection with a single-threaded connection, except on an iperf to a server on a high quality 10gig/sec link.

I notice that on the speed.hyperoptic.com multi-threaded (Ookla-based) tester, the download and upload speed test do that nice curve almost before stabilising at a pretty solid speed. If you're on the 150 Mbit/sec package, that shows you're able to get maximum achievable throughput at least, plus a bit extra to accommodate TCP overhead (how generous of them).

The graphs look like you're getting an initial burst of higher throughput before it settles down. 167 Mbit/sec is actually about 20 MB/sec, 150 Mbit/sec is 18.75 MB/sec (including overheads). The brief peak of 24 MB/sec to me looks like a calculation error.

If you do a single-thread download in your browser from one of the following test servers while monitoring your bandwidth (I like NetMeter Evo), how does it graph?

http://ams-speedtest-1.tele2.net/1000GB.zip
http://speedtest-ams2.digitalocean.com/1gb.test
http://ipv4.download.thinkbroadband.com/1GB.zip

http://ipv6.download.thinkbroadband.com/1GB.zip
http://176.67.170.244/speedtest.256mb (Manchester), http://109.123.87.183/speedtest.256mb (London)

As Caged mentioned, hook straight in to your Hyperoptic socket and test also. If the Hyperhub is misbehaving, you've eliminated it. Which model Hyperhub do you have? The Tilgin or one of the two ZTEs?


I haven't tested on the slower Hyperoptic packages, always had the gigabit. However this might just be normal behaviour for your connection. I presume lower bandwidth packages are done by HO using policer queues (or traffic shaping) on the switch your socket connects to. If so, that initial spike of higher bandwidth (smoothly curving off to your average speed) may actually be normal and a characteristic of your traffic as the speed settles down to the committed rate. The traffic management policy may permit a brief spike of higher speed before the policies take effect.



A way to test your true burst and sustained bandwidth would be to do repeated iperf3 tests to and from a public iperf server. (see https://iperf.fr for publicly accessible iperf3 servers and instructions on use). If the speed tails off in the same way, it's probably just a characteristic of how Hyperoptic manages the line speed.

This is why I suggest making a Linux LiveUSB stick and booting from that for tests, it totally eliminates Windows from the equation. I found my NIC and overall network performance was way better under Linux. Better network stack, fewer overheads.


Other possibilities are that your building's incoming bandwidth is almost at capacity and you're being locally traffic managed for certain classes of traffic. I believe in some installations with high numbers of customers sometimes bandwidth suffers a bit at peak time, but then you'd probably notice that on all types of usage.

During a long period of downloading or uploading, with a speed graph app (like NetMeter Evo I mentioned above) watch the speeds. If you get sudden drops to almost zero then a recovery back up to a normal speed, that's a possible problem with your CPU setup or drivers/DPC/packet loss. Less likely is something physical like a dodgy patch to the Hyperoptic's upstream network switch. It may also potentially be contention on the incoming link supplying the whole building.

If you're running Windows 10, also install LatencyMon and run it while doing a heavy download. If you get significant DPC latency issues and warnings on screen, it's your PC unable to keep up for some reason, usually a driver-related issue.



Slightly related -
I did a bunch of testing to see what would minimise OS and driver impact on network throughput. On my network adapter, for better overall performance (this is what Intel and a few research organisations recommend for more reliable high-speed throughput), disabling the offloading features of your network adapter can reduce latency and improve performance (at a possible hit to CPU usage). This is in the NIC's Advanced tab...

Make a note of what settings you were using before changing so you can revert if no difference! This is how I've configured my NIC, an onboard Intel I218-V:
(settings not mentioned are inconsequential or left at defaults)

  • Adaptive Inter-Frame Spacing: enabled
  • Flow Control: Rx & Tx Enabled
  • Interrupt Moderation: Enabled
  • Interrupt Moderation Rate: Adaptive
  • IPv4 Checksum Offload: Disabled
  • Large Send Offload V2 (for IPv4) and (IPv6): Enabled (...though some people recommend Disabled, I found it slightly increased my CPU usage)
  • Receive Buffers: as high as possible (I can get to 2048, you may only be able to do 512 or 1024)
  • Transmit Buffers: as high as possible
  • TCP Checksum Offload: Disabled (for IPv4 and IPv6)
  • UDP Checksum Offload: Disabled (for IPv4 and IPv6)
I run the 29/09/2017 drivers (core driver version 12.17.8.7, PROWin v22.10) because later revisions appeared to be causing massive DPC latency problems. Stuttering, audio issues, momentary freezes etc.

I don't think changing these settings on your card will make much if any difference. But thought it worth making a note for others on the gigabit package who might be going through the same testing I did.



Ha yeah the 'DIY' is still on going mate, as is playing life size Tetris as we find places for things. Settling in though, thanks for asking. Also thanks for taking the time to provide the detailed information. This is all new to me so is interesting to soak up.

I am on the 150mb package, anything bigger seemed overkill really. Coming from sky Fibre MAX I was getting about 75mbs. So this is great now.

The Router is a ZTE ZXHN H298A (https://postimg.cc/gallery/1h3lqwxj0/)

I downloaded NetMeterEvo and ran a few tests. I tried to do a few tests per session but then started to get blocked by the browser, might be a windows explorer security protocol (pretty new to Win10). I imagine due to me repeating the tests back to back?

I basically performed the tests via the router and then plugged direct into the socket. It was obvious straight away that plugging into the sockets yielded perfect download speeds. So potential Router issues or something within its set up needs adjusting? I took a screen of two games used for testing purpose during D/L. It was solid, nothing like when it is utilising the Router, which is erratic.

Tests Via Router
https://postimg.cc/gallery/1cgc76bi4/

Tests Direct into the socket
https://postimg.cc/gallery/1prj5zem4/

Sorry the tests aren't consistent and not all of them were conducted. I can try and do them next week. Had a busy weekend and wanted to ensure no one else was using the bandwidth (the missus is a serial Netflix user :D).

So do you think something within the Router set up is causing the issue?

Many Thanks

B
 

~cw

~cw

Associate
Joined
2 Jan 2019
Posts
164
Those speed tests all look very consistent, it doesn't look like you're suffering any speed loss! You've got the same router as me and I've noticed no slowdowns caused by the router on the gigabit package. I also do numerous port forwards (which all eat into CPU usage on the router) and it doesn't cause any slowdown. I think the ZTE has hardware acceleration for gigabit packet switching as well, so I would expect it to run at full speed with no problems.

When going through the router, any initial spike you're seeing is possibly just due to the router having to set up the session, do port address translation and talk to your computer (whereas when you connect direct your NIC is talking to the Hyperoptic switch in your building). The router may be buffering the initial sent packets then delivering as fast as possible once it's ready to handle the connection, which might be causing that initial 'blip' of higher speeds. Once the router settles down, it can transmit the data at wire speed for the rest of the download, so you notice your speeds immediately return to normal and stay pegged at the true maximum speed.

On the router, you said Firewall you'd tried on various settings - I run mine on Low and leave "anti-hacking" disabled (basically useless). in Filtering, I have everything disabled, DMZ disabled and IPSec switch disabled.

In NetMeter, the very thin spikes (like in https://postimg.cc/K4LpFf47) I'm 99.9999% certain are just Windows or the app misreporting bandwidth utilisation for that instant. Unfortunately this skews the graph maximum value, and the axis lines are not absolute values, so I can't see what the actual speed is measured as. Re-run the tests and take screenshots while they're happening, not afterwards, then you can see the momentary speeds in NetMeter.

The trick with NetMeter is to watch the second-by-second upload / download speeds while testing (in the bar below the graph). Run a Steam and file download test both when direct into the socket and via the router, I believe you'll see your speeds are pretty much the same (excluding that very brief burst at the start when going via the router).


I wouldn't worry about doing speed tests while people are on Netflix, the router will be able to handle both fine. You'd possibly just notice your speed test dipping slightly as new chunks were downloaded by the Netflix player.

Having a very generous upstream speed - you also get (up to) 150 mbit upload - really helps avoid congestion and buffering. You don't typically suffer the same symptoms of overloaded connections that were common with ADSL and slower cable packages (where you might have 16 or 20 megabits down but only half a megabit up, very quickly that can become saturated just from the packet acknowledgements TCP sends for every received packet).


If the spike isn't caused by the router, it might be due to the PC being very briefly too busy to adequately process bandwidth usage in real time, this can happen when another process requests a CPU interrupt (which then skews Windows' own calculations) - all sorts of possibilities. That's why watching speed over a longer average and filtering out the 'noise' gives you the accurate reading. However in your case, when you go direct you don't get the spike, so that points me back towards the router as being the 'cause', though in truth you're not losing any speed that I can tell (I'd need more bandwidth numbers to confirm though).

Reason I think your speeds are fine is because in the one 'test via router' screenshot you took while a download was actually running, the speed at that moment was almost 20 MB/sec, which is normal. Better than the quoted package speed still. https://postimg.cc/K4LpFf47 Run that 1000GB test again a couple of times via router and via direct connection, and expand the graph to show a longer period of time - chances are you'll see the same brief spikes and dips due to upstream congestion which is fairly normal.

Though you've not mentioned speeds during the tests, all the netmeter graphs during file download tests are basically steadily flat across the top, as you'd expect on a line able to max out at full speed. The steady 20 MB/sec is actually about 160 Mbit/sec so you're doing better than the advertised package speed.



I've occasionally experienced a speed test which won't run, usually it's the browser. Typically I can download files or run speedtests repeatedly without issue. It's usually a browser issue if a web-based speed test won't run. Use Chrome in Incognito or flip-flip between Firefox and Chrome, you'll notice as you add extensions or plug-ins that speed may suffer as well, so best to run them in a completely fresh browser with zero extensions.

I had good results with a fresh Firefox install when Chrome started losing about 10% of my speedtest speeds, eventually realised one particular extension was causing Chrome to internally lag and lose efficiency.


If your console or other device is not always achieving full speed, unfortunately, welcome to ultrafast internet. The ISP's network is capable of doing full speeds all the time, but (as they've said to me) as soon as your data passes over the internet, all bets are off. Basically they're not paying for 100% guaranteed maximum bandwidth throughput for every customer from their upstream providers at all times, understandable due to the economics of it. However as a result the providers (who also make their own decisions about how to route traffic through their network) sometimes end up compromising our maximum real world available speeds. Not really anything we can do to improve this, unless you have a server on the far end which also gets its connection via one of those providers, then you can complain to the carriers as a direct customer. Not possible for most people though.


You also notice that many remote systems are incapable of running at your connection's max speed, especially when you start to get above 100 Mbit/sec internet connections. In some cases, servers won't be able to serve content as fast as your connection, or they might be throttling their own speeds to serve multiple users.

Add to this upstream peering agreements your ISP has with 'transit providers' (bandwidth providers to other networks), available bandwidth capacity inside the third party provider networks, plus overall 'busyness' of the systems all contribute to slower than expected speeds. Many factors.


It may be that the XBox downloads from Microsoft's CDN using single-thread downloads (vs. multi-thread downloads like Steam, Ubisoft and others do), at which point you'll notice your speed fluctuating or not maxing out, because it's rare that a single threaded TCP session can utilise the maximum theoretical bandwidth of a link, especially when it has to travel via the Internet taking weird routes through several networks. Nothing you can do about this really.


Am I missing anything else, have you noticed bad speeds at certain times or to specific things, or does this explain what you're seeing? Forgive me if any of it seemed condescending, difficult to know sometimes how much experience people have in networking or the weird ways TCP behaves sometimes.

(Obviously these are only my personal opinions and views as a punter, I have nothing to do with Hyperoptic except they take my money each month for internet)
 
Soldato
Joined
11 Jan 2016
Posts
2,563
Location
Surrey
Didn't realise my contract was coming to an end but Hyperoptic sent me a reminder with an offer of £40.50 a month to keep my 1000/1000. Very nice of them and I obviously took it. Had two full years with them since moving to my new build house at £45 a month. Could not go back to slow internet.
 
Associate
Joined
7 Oct 2013
Posts
1,302
Didn't realise my contract was coming to an end but Hyperoptic sent me a reminder with an offer of £40.50 a month to keep my 1000/1000. Very nice of them and I obviously took it. Had two full years with them since moving to my new build house at £45 a month. Could not go back to slow internet.
Nice! Via post or email? I must be coming up for renewal soon too
 
Back
Top Bottom