ISP Traffic Manipulation policies

Forums › New Zealand Broadband › ISP Traffic Manipulation policies

1 | 2

raytaylor

1602 posts

Uber Geek

Trusted

Subscriber

Reply # 666291 2-Aug-2012 15:19
freitasm: I have no knowledge of ISPs in New Zealand making this kind of information public. One of the sysadmins at orcon? i think a few years ago did an authorised two-part blog about how they manage traffic. I cant remember exactly which ISP but i am pretty sure it was orcon. Ray Taylor Napier, NZ www.taylorcommunications.co.nz

mercutio

781 posts

Ultimate Geek

Reply # 666439 2-Aug-2012 18:38
Talkiet: Heh, Of course I'm not going to comment on a commercially sensitive topic as current and future bandwidth capacity, EVEN if I knew the up to the minute figures. What I will say is that my plain vanilla (nothing different from any other Telecom Retail customer) connection at home achieves the expected speeds 24*7, within the constraints of what TCP does with the latency and remote server capacity. Of course peering both nationally, and internationally can influence latency and "remote server capacity". And the choice of provider can lead to congestion anywhere along the path. d). Governments put public money in to improving "local" bandwidth without addressing the "next hop" problem (which really only benefits the big content providers) The government money is going to benefit everyone, not just through UFB but through RBI as well. The "next hop" problem can't be solved unless you can figure out how to make light go faster than it does already, or some cunning way to drill straight through the earth rather than around it :-) ... CDNs and Proxies don't just need to benefit the big content providers either. For example, my personal website (www.nzsnaps.com) is on a US server and for most people in NZ is served off a CDN substantially closer than that. I don't see how the government money is going to benefit me? Maybe it'll benefit a lot of people, but I can't see it benefitting me right now, nor in the near future. I'm not in a fibre area. I'm not likely to get faster international speeds. I'm not likely to get a "better level of service" in any way really. That site didn't seem to load particularly quickly for me. It's slower than say www.anandtech.com. The problem with cdn's is that often content isn't fresh in cache, and still has to be requested from overseas. If you want to have it preloaded you have to pay more money, and things like akamai are going to charge an earth compared to companies like cloudflare that host their own servers. That said, I run all my browsing through a squid proxy, and see the time that requests take to load - and photos.smugmug.com and cdn.smugmug.com are fast and using akamai, and www.nzsnaps.com is using akamaiedge and slow (over 1 second per request) I don't disagree that cdn's can help performance - but I find that often performance can still be subpar. And I do strongly believe that push-caching is better than pull-cachcing. BTW can we leave the guessing of what I use my bandwidth for to my ISP... I could care less what you (or anyone else) use your bandwidth for, and it's none of your business (but I think you'd be surprised ;P). Completely fair comment - but if I had a buck for every time someone told me they needed to download Linux ISOs every day, I'd have more money than Bill Gates - mostly cos he'd be bankrupt cos everyone was using Linux. It wasn't a dig at you at all - it was my generic tongue in cheek term for any large download. One thing I've noticed - is that Linux ISO's tend to download fast. I think the kind of people who are into Linux like to seed, start mirrors, and share content. The problems with performance I usually find are to shopping web sites (i expect medium, some are downright slow), news web sites (i expect instant), forums (i expect high performance), and uploading. The problem of slow web sites, seems to often be the origin server rather than cdn's. It seems common for searching to be slow, or to have minor delays here and there with sites. But sometimes sites seem to get in a mood and get slow. News web sites are generally good, but some US ones can be kind of slow. Especially when they use akamai. (when I've looked at where things are going) bbc tends to be fast - cnn and washington post tend to be slow. Forums - well, whirlpool (austraila) and hardforum (us) both seem to be considerably faster than geekzone (nz) for me, and yet they're further away. I think the mistake some people make is that they I assume cdn's will fix everything. And that having well setup/optimised site and fast database helps. Curiously both of those fast web sites are using nginx, which is on the more technical direction rather than just throwing up a page on shared hosting, and have probably done other optimisations too. But it shows more than a cdn. Because the pages load faster. And upload speeds? Well, New Zealand seems to be doing very little to implement annex-m, so that common users can at least choose to get a little boost in upload speed. The vast majoriy of users are stuck on sub-1 megabit upload speeds. But right now as it stands, it appears 3g is the easiest option if users want to get upload speeds over 1 megabit. Which is crazy! If I had higher upload speeds, I could consider saving a lot of my files remotely. But even just sending a 300k screenshot takes seconds. And TBH, I don't even think 10 megabit is fast enough for home users to really enrich their experience. And 10 megabit is as fast as VDSL is capped at - and the low end fibre plan.

mercutio

781 posts

Ultimate Geek

Reply # 666440 2-Aug-2012 18:39
raytaylor: freitasm: I have no knowledge of ISPs in New Zealand making this kind of information public. One of the sysadmins at orcon? i think a few years ago did an authorised two-part blog about how they manage traffic. I cant remember exactly which ISP but i am pretty sure it was orcon. Do you mean: http://www.orcon.net.nz/lifestyle/page/broadband_service_control_in_the_orcon_network/ ?

raytaylor

1602 posts

Uber Geek

Trusted

Subscriber

Reply # 666929 3-Aug-2012 13:00
Yes thats exactly it. Ray Taylor Napier, NZ www.taylorcommunications.co.nz

Publius

164 posts

Master Geek

Reply # 667123 3-Aug-2012 17:33
The lower latency of fibre (2ms) over dsl (15-25ms) also helps too in that tcp will ramp quicker from start and after a dropped packet.

Talkiet

1405 posts

Uber Geek

Trusted

Telecom NZ

Reply # 667137 3-Aug-2012 17:48
Publius: The lower latency of fibre (2ms) over dsl (15-25ms) also helps too in that tcp will ramp quicker from start and after a dropped packet. The DSL latency is due to various depths of interleaving being enabled. While fibre will pretty much always be lower, it's almost never necessarily anything like 13-23ms worse. As an example, the RTT from my desktop machine to the first hop outside my router is about 7ms Cheers - N

insane

1275 posts

Uber Geek

Trusted

Reply # 673518 16-Aug-2012 22:48
mercutio: raytaylor: freitasm: I have no knowledge of ISPs in New Zealand making this kind of information public. One of the sysadmins at orcon? i think a few years ago did an authorised two-part blog about how they manage traffic. I cant remember exactly which ISP but i am pretty sure it was orcon. Do you mean: http://www.orcon.net.nz/lifestyle/page/broadband_service_control_in_the_orcon_network/ ? There was later another follow up which is here : http://www.orcon.net.nz/lifestyle/page/broadband_service_control_in_the_orcon_network_followup/ However both of those articles are probably no longer relevant as they have for at least two years now been using a different appliance to manage traffic. ie, Sandvines which incidentally they rate rather highly.

mercutio

781 posts

Ultimate Geek

Reply # 673523 16-Aug-2012 23:10
Talkiet: Publius: The lower latency of fibre (2ms) over dsl (15-25ms) also helps too in that tcp will ramp quicker from start and after a dropped packet. The DSL latency is due to various depths of interleaving being enabled. While fibre will pretty much always be lower, it's almost never necessarily anything like 13-23ms worse. As an example, the RTT from my desktop machine to the first hop outside my router is about 7ms Cheers - N one thing to note is that "ping" tests don't do full sized packets - which is likely to push the latency more towards publius's numbers - some people think they don't upload much - but just going to web pages, requesting new pages, and acknowledging packets can use some bandwidth. with linux default settings it's extra high for upload usage - it does something called quickack where the first 16 packets will be acknowledged straight away instead of joined onto the next one. And linux also enables tcp/ip timestamps by default which also increase overhead. because adsl is typically 600 to 1100 kbit the amount of bandwidth usage on upload is quite scarce, and small amounts of congestion can happen easily on the upload channel with even casual use.

mercutio

781 posts

Ultimate Geek

Reply # 673529 16-Aug-2012 23:25
insane: mercutio: raytaylor: freitasm: I have no knowledge of ISPs in New Zealand making this kind of information public. One of the sysadmins at orcon? i think a few years ago did?an authorised two-part blog about how they manage traffic. I cant remember exactly which ISP but i am pretty sure it was orcon. Do you mean:?http://www.orcon.net.nz/lifestyle/page/broadband_service_control_in_the_orcon_network/?? There was later another follow up which is here :? http://www.orcon.net.nz/lifestyle/page/broadband_service_control_in_the_orcon_network_followup/ However both of those articles are probably no longer relevant as they have for at least two years now been using a different appliance to manage traffic. ie, Sandvines which?incidentally?they rate rather highly. no-one appears to be publishing anything recent. it's curious that orcon say they're using 10 gigabit on transit links. 5 10gigabitethernet2-1.core1.lax1.he.net (72.52.92.121) 9.983 ms 10.440 ms 9.920 ms 6 ge-5-0-1.cre1.la1.odyssey.net.nz (206.223.123.92) 1.217 ms te3-4.mpd01.lax05.atlas.cogentco.com (154.54.3.142) 0.845 ms ge-5-0-1.cre1.la1.odyssey.net.nz (206.223.123.92) 0.838 ms 7 121.99.12.0 (121.99.12.0) 134.819 ms xe-0-1-0.cre1.la1.odyssey.net.nz (38.104.84.30) 0.576 ms 1.103 ms Their router seems to say ge. Although it does seem to say xe for cogent. That path seems mental - alternating between he.net/cogent. But tracing to other ip's it was just going through cogent. Anyway, both he.net and cogent are known for having issues with congestion on their network reasonably regulalry - and have network quality such that it's generally suggested to use as a tertiary connection, not even as a secondary. And in effect low quality providers overseas can bottleneck internet experience even if the traffic isn't shaped at the NZ end. This is one of the reasons that west coast speed tests tend to be faster.

1 | 2