Why you don't get the whole 100mbps in UFB and Why it's capped to 100mbps only??

Forums › New Zealand Broadband › Why you don't get the whole 100mbps in UFB and Why it's capped to 100mbps only??

1 | 2 | 3

507 posts

Ultimate Geek

#833430 10-Jun-2013 10:44

At end of day my fibre connection gives me lower ping on the games I play, its about the same cost as my DSL was and Snap seem good to deal with.

Thumbs up from me thus far for UFB

I think its capable of more but meh.

I am concerned though as more people get it locally performance will drop to worse than DSL - is this possible?

The force is strong with this one!

mercutio

1392 posts

Uber Geek

#833437 10-Jun-2013 10:55

rossmnz: At end of day my fibre connection gives me lower ping on the games I play, its about the same cost as my DSL was and Snap seem good to deal with.

Thumbs up from me thus far for UFB

I think its capable of more but meh.

I am concerned though as more people get it locally performance will drop to worse than DSL - is this possible?

it's possible, but unlikely. the main concern with dsl congestion was that handovers were atm (~155 megabit) then 1 gigabit both of which could get full.

with pricing set at something like $100/month for gigiabit, and $300/month for 10 gigabit for handovers the main thing holding back 10 gigabit adoption will be the high cost of line cards.

webwat

2036 posts

Uber Geek

Trusted

#841338 21-Jun-2013 23:05

mercutio:
wired: With the same number of people on a PON fibre with faster plans there is more chance of two people downloading at the same time and therefore congesting the network even if it is only briefly. When congestion occurs, there is a chance that packets will be dropped because the queues are not big enough to handle the congestion.

The government has specified 100 Mbps plans with a maximum of 2% packet drop. With only 2.5 Gbps available in the DS, the PIR of the plans needs to be such that enough people can use the same shared fibre to get the $ benefits of PON while avoiding congestion and therefore packet drop. If the 2% packet drop limit was got rid of, then the LFCs could allow more congestion in their networks and hence probably offer faster plans.

you're not going to get 100 megabit transfer speeds with 2% packet loss though. i'd rather some kind of round robin scheduling between the ports myself, and to just have gigabit.

2.5 gigabit to 24 users is heaps, but i expect there'll only be gigabit backhaul generally to keep costs down.

i'm not quite sure how there's more chance of two people downloading at the same time though. given the same usage limits and faster connections there should be less chance of two people downloading at the same time because things download quicker.

anyway, for the most part it's a con. the vast majority of users won't want to pay for expensive internet, and it'll mean that copper is deprioritised for the majority of customers staying on copper be it for cost reasons, renting, or lack of availability.

When congestion occurs, the downloaders should slow down their requests for so much data according to their TCP receive window.

The OLT (its not a headend) does have a sophisticated sharing regime equivalent to round robin, but would generally be part of a chassis with multi gigabit backhaul. How much contention happens upstream from the OLT card is another subject that government doesn't seem to have any interest in, so in my view it was pointless for CFH to focus on limiting the PON split ratio without setting up the system to produce competition between Layer 2 service providers.

Time to find a new industry!

Glazza

279 posts

Ultimate Geek

Trusted

#841353 22-Jun-2013 00:01

We operate a Fibre Network, using the 7330 equipment, much like some of the LFC's.

We have developed a couple of profiles, one with aligns itself closely to the TCF specification, where latency will always be less than 5ms, this profile basically has a hard limit when you reach bandwidth limits.. ie 100Mbps, and therefore unless your service provider is doing rate shaping ahead of us, you end up with a saw tooth looking throughput graph

We have another profile, which we provide more generous MBS and CBS buffers - the advantage, you get very close to a clean 100Mbps, which on the throughput graph looks nice and smooth, the downside, because of the buffering, if you are pushing the limits of the connection, and pushing large packets, we can not commit to a latency of 5ms.

With regards to the maximum of 100Mbps, G-PON will allow an allocation of an entire PON bearer to a single customer - so 2.4Gbps down, and 1.2Gbps up.

Typically, a B class SFP has been designed for a 32 way splitter in mind - the limitation been the loss that is introduced at the splitter. So when operating with 32 ways splitters, if you have a fully utlised PON bearer, and all the connections on that bearer are fully utilising there connection, the maximum throughput to each customer would be limited to 75Mbps.

There was a reference before with regards to the change of 'two people downloading the same thing at the same time was minimal' - and even if there were too, each customers connection is encrypted by the OLT, and decrypted by the ONT - so even your ONT is receiving the data for all the customers on the same PON - it can only decrypt the data destined for it.

jarledb

Webhead
3257 posts

Uber Geek

Moderator

ID Verified

Trusted

Lifetime subscriber

#841366 22-Jun-2013 02:09

raytaylor: To be honest, if someone is moaning about only getting 95mb on a 100mbit connection - they are probably the same people that complain about their 160gb hard drive only having 148gb of usable storage capacity

We might understand why, but for a regular consumer there is no way of knowing that a 160 GB HD won't actually give you 160 GB of space to store stuff. The same goes for broadband.

Where I live at the moment my connection is sold as 70/10 Mbit/s and that is more or less what I get (sometimes even higher than that speed). And that is how broadband is sold here. HDs are still sold the same way as in NZ though...

Jarle Dahl Bergersen | Referral Links: Want $50 off when you join Octopus Energy? Use this referral code
Are you happy with what you get from Geekzone? Please consider supporting us by making a donation or subscribing.

mercutio

1392 posts

Uber Geek

#841444 22-Jun-2013 11:52

jarledb:
raytaylor: To be honest, if someone is moaning about only getting 95mb on a 100mbit connection - they are probably the same people that complain about their 160gb hard drive only having 148gb of usable storage capacity

We might understand why, but for a regular consumer there is no way of knowing that a 160 GB HD won't actually give you 160 GB of space to store stuff. The same goes for broadband.

Where I live at the moment my connection is sold as 70/10 Mbit/s and that is more or less what I get (sometimes even higher than that speed). And that is how broadband is sold here. HDs are still sold the same way as in NZ though...

my cheap 120gb samsung 840 ssd says up to 530mb/sec, and goes at 545mb/sec on benchmark.

the simple solution would be to sell ufb connections at up to 25/8 and 80/40

webwat

2036 posts

Uber Geek

Trusted

#841794 22-Jun-2013 22:53

mercutio: my cheap 120gb samsung 840 ssd says up to 530mb/sec, and goes at 545mb/sec on benchmark.

the simple solution would be to sell ufb connections at up to 25/8 and 80/40

But since that 25/8 or 80/40 would be an estimate based on theoretical throughput of certain kinds of traffic and usage scenarios under specific test conditions, would the marketing have to specify how the advertised speed was measured? Its quite established to advertise the headline speed and everybody knows they have to expect actual throughput to be dependant on how they use it.

Time to find a new industry!

Tradepub

Free professional, reference and technical white papers (affiliate link).

mercutio

1392 posts

Uber Geek

#841799 22-Jun-2013 23:03

webwat:
mercutio: my cheap 120gb samsung 840 ssd says up to 530mb/sec, and goes at 545mb/sec on benchmark.

the simple solution would be to sell ufb connections at up to 25/8 and 80/40

But since that 25/8 or 80/40 would be an estimate based on theoretical throughput of certain kinds of traffic and usage scenarios under specific test conditions, would the marketing have to specify how the advertised speed was measured? Its quite established to advertise the headline speed and everybody knows they have to expect actual throughput to be dependant on how they use it.

well the speed is going under ethernet speed.

on gigabit ethernet at home with 1500 mtu at home i'm getting 949 megabit/sec tcp between windows/linux.

so 100 megabit ufb should be able to do ~94 megabit. but people are getting 90 and below from what i see.

some guy on nzherald was saying that his 30/10 ufb was going at 20/5, although he did seem to think that was ok.

ethernet speeds are reasonably well known. it's not unusual to see 930 megabit/sec on gigabit on lan. although 940 and above tends to mean you have to have tcp time stamps disabled, good ethernet cards etc. (although one of my ethernet cards is onboard realtek, hmm.. )

i'd rather see ufb connections that are provided at much faster speeds, and then the isp decides on the speed. it seems at present all the isp's are doing "max" speed connections, but if it was normal for a connection to be 250 megabit or something, then the isp can choose to do 30, 50, 100, 250 etc on that it'd be preferable to my mind. then can start having things like up to 30 megabit international up to 200 megabit national etc.

although i suppose isp's could already start shifting to rate limited international on 100 megabit plans to ensure that users get good speeds consistently rather than wavering because a whole lot of users suddenly start torrenting some popular huge file or something that just comes out.

jarledb

Webhead
3257 posts

Uber Geek

Moderator

ID Verified

Trusted

Lifetime subscriber

#841800 22-Jun-2013 23:03

webwat: everybody knows they have to expect actual throughput to be dependant on how they use it.

Both my wife and I are geeks, and my wife didn't know that. So everybody is a false statement.

There is no way to in real live actually get the speeds that are marketed for internet connections in New Zealand. Its the reason they changed how its done here in Norway, and I think its a good reason to change it in NZ as well.

The consumer should at least be able to expect to get something near the speeds that the packages are sold as. The costumer should not have to worry about the overhead of the connection etc.

mercutio

1392 posts

Uber Geek

#841801 22-Jun-2013 23:05

jarledb:
webwat: everybody knows they have to expect actual throughput to be dependant on how they use it.

Both my wife and I are geeks, and my wife didn't know that. So everybody is a false statement.

There is no way to in real live actually get the speeds that are marketed for internet connections in New Zealand. Its the reason they changed how its done here in Norway, and I think its a good reason to change it in NZ as well.

The consumer should at least be able to expect to get something near the speeds that the packages are sold as. The costumer should not have to worry about the overhead of the connection etc.

both citylink ethernet and telstraclear cable give tcp/ip throughput of the plan speed.

mercutio

1392 posts

Uber Geek

#841802 22-Jun-2013 23:12

another thing to keep in mind that speedtest.net isn't accurate at gigabit speeds.

it won't even show a result link for speeds above 1 gigabit.

insane

3239 posts

Uber Geek

ID Verified

Trusted

#841803 22-Jun-2013 23:13

mercutio:
the simple solution would be to sell ufb connections at up to 25/8 and 80/40

That would never work because what ISP marketing department is going to advertise a 80/40 connection when the next is advertising 100/50.

Also if you did a UDP test with say JPERF to a suitable host you'd max the link out, or at least get 99/49.

It's similar to drinking coke from a can, sure it might be a 355ml can but unless you crack that tin open and lick that thing out you're not getting 355ml out of there!

jarledb

Webhead
3257 posts

Uber Geek

Moderator

ID Verified

Trusted

Lifetime subscriber

#841819 22-Jun-2013 23:53

insane:That would never work because what ISP marketing department is going to advertise a 80/40 connection when the next is advertising 100/50.

There are a couple of ways to fix that.

1) The ISPs agree amongst themselves to advertise the speeds more realistically.

2) Commerce commission force the ISPs to use a more realistic way of advertising speeds. With a common way of measuring that does not exclude overhead.

mercutio

1392 posts

Uber Geek

#841823 23-Jun-2013 00:57

insane:
mercutio:
the simple solution would be to sell ufb connections at up to 25/8 and 80/40

That would never work because what ISP marketing department is going to advertise a 80/40 connection when the next is advertising 100/50.

Also if you did a UDP test with say JPERF to a suitable host you'd max the link out, or at least get 99/49.

It's similar to drinking coke from a can, sure it might be a 355ml can but unless you crack that tin open and lick that thing out you're not getting 355ml out of there!

you sure it can do 99 megabit udp?

i want some ufb connections to experiment with.

normal overhead of tcp without timestamps is 40 bytes per packet, with 1500 mtu that's 0.026% overhead. normal overhead of udp is 28 bytes per packet, with 1500 mtu that's 0.018% overhead, which should equate to 97.3 megabit, with tcp/ip and 98.1 megabit with udp. the difference isn't likely to be that noticable is it?

on top of that, when testing with udp, if you want something that actually is like real world data, you'll need a sequence number at the very least, as packets may be delivered out of order.

also jperf is like the windows iperf? windows iperf doesn't deal well with udp at high bitrates linux works a lot better. iperf itself uses lots of cpu, i think the implementation isn't very efficient. it's ok on windows with tcp/ip though..

Talkiet

4792 posts

Uber Geek

Trusted

#841824 23-Jun-2013 01:00

mercutio: another thing to keep in mind that speedtest.net isn't accurate at gigabit speeds.

it won't even show a result link for speeds above 1 gigabit.

speedtest.net is an estimate of peak speeds at the best of times.

Go read how they calculate the numbers they present you...
Not saying they don't usually get it close to POTENTIAL peak speed, but it's NOT a measure of volume over time.

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.

1 | 2 | 3