[AusNOG] TPG Peering

John Lindsay johnslindsay at mac.com
Wed Oct 22 08:58:51 EST 2014


So make it clear where the problem is and the end users can make their choice of provider. 

John Lindsay

> On 22 Oct 2014, at 8:15 am, Jared Hirst <jared.hirst at serversaustralia.com.au> wrote:
> 
> I agree with this, the issue that I have is that we are sending 500mbits of traffic to hundreds of users, and it is mainly gaming traffic AND TV streaming, therefore the need for them to have it ASAP and with low latency is very important to us. If I was just providing a CDN for downloads then there would be no cause to worry about the 4 hours of congestion, but this is for live games and TV.
> 
>> On Wed, Oct 22, 2014 at 6:59 AM, Mark ZZZ Smith <markzzzsmith at yahoo.com.au> wrote:
>> 
>> 
>> 
>> 
>> 
>> >________________________________
>> > From: Jared Hirst <jared.hirst at serversaustralia.com.au>
>> >To: Bob Woolley <boblobsta at gmail.com>
>> >Cc: "ausnog at ausnog.net" <ausnog at ausnog.net>
>> >Sent: Tuesday, 21 October 2014, 20:41
>> >Subject: Re: [AusNOG] TPG Peering
>> >
>> >
>> >
>> >Because generally they wont direct peer. We are trying to send them 500+ mbits though and when it's congested this is causing issues for us and their users.
>> >
>> >
>> 
>> Is it really causing issues for their users?
>> 
>> One thing to bear in mind is that TCP attempts to utilise what ever bandwidth is available at the time, which is quite reasonable as a design goal, because the network has been "100% paid for" so why not try to use 100% of its capacity when ever possible? Multiple TCP connections over the same path should back off such that the available bandwidth is roughly equally shared between all of them.
>> 
>> However, when it comes to what is actually being delivered over that TCP connection, aiming for 100% network utilisation may not be necessary when the data is being delivered for a human. If the data is 'bulk' data e.g., a large file download, once the total time it takes to download goes past a human's longer term focus attention span, IOW, the time that they're focus on watching it occur, then the total time it takes is less important. Consequently, allowing congestion to occur over paths that these types of TCP connections are taking isn't necessarily as bad as it appears. The only consideration then would be how is congestion impacting latency sensitive applications also operating over that congested link. If you have none of those, then the congestion is far less harmful, as long as the large downloads arrive within a reasonable time (and a reasonable time seems to be in multiples of hours).
>> 
>> For example, if the recent IOS 8 update was going to take 45 minutes to download over a congestion free link, but 60 to 75 minutes over a congested link, I don't think many if any people would either care and notice.
>> 
>> I don't know of studies that show where the 'focus attention switch' threshold is for large downloads, however I think it is only in the order of a few minutes - the time where you start the download and then switch away from observing it and start doing something else e.g., start web browsing, making a coffee, got to watch TV etc. If there isn't a prediction shown as to how long it takes, then I think people tend to check every 5 to 10 minutes, and perhaps increase their checking interval if progress isn't significant e.g., start checking once every 15 minutes, then 30 etc. After that, if progress still isn't quick enough, then people will leave it running without checking for many hours, and perhaps overnight.
>> 
>> I think this further shows that latency is more important to humans than bandwidth/throughput. IOW, we desire instant gratification, or as near to it as we can achieve or pay for, and continue to be motivated to get closer to instant gratification (the cliché "patience is a virtue" is fundamentally advice to try to suppress these instant gratification desires).
>> 
>> People who want high bandwidth/throughput so their downloads have short completion times are actually still saying latency is more important to them - they want lower *latency*[1] downloads. Whether that is a realistic expectation or not depends on how much they're willing to pay. As TPG customers would typically be more price sensitive, it would be reasonable for them to have to accept higher latency "anything" than those of other higher price/quality ISPs.
>> 
>> So in summary, if the content you're sending TPG is 'bulk', then congesting over links towards them may not be as much of a problem as it appears. If it is a problem, TPG's customers will or should be ringing the TGP helpdesk to complain about it. That is where the cost of this congestion is to TPG. If TPG don't care about their customers, those customers will find a better ISP. If TPG do care about those customers, then TPG should be more willing to come to an arrangement (which might include private peering) that eliminates the congestion associated helpdesk costs.
>> 
>> [1] https://en.wikipedia.org/wiki/Latency_(engineering)
>> 
>> >
>> >
>> >
>> >On Tue, Oct 21, 2014 at 7:27 PM, Bob Woolley <boblobsta at gmail.com> wrote:
>> >
>> >Why should that be a long shot?
>> >>because tpg don't like networking with others
>> >>
>> >>
>> >>- Bob W
>> >>
>> >>
>> >>
>> >>On 21 October 2014 18:18, Mark Newton <newton at atdot.dotat.org> wrote:
>> >>
>> >>
>> >>>
>> >>>On Oct 21, 2014, at 10:50 AM, Jared Hirst <jared.hirst at serversaustralia.com.au> wrote:
>> >>>
>> >>>Looking to talk to someone about peering direct with TPG (I know it's a long shot) but the issue is that everday between 6-9pm we see congestion to TPG, upon looking at their PIPE port we can see it's maxed at 20 gbits, and hence this would be causing our issues.
>> >>>
>> >>>Why should that be a long shot? Sounds like precisely the kind of thing that’d justify a private peering relationship.
>> >>>
>> >>>
>> >>>  - mark
>> >>>
>> >>>
>> >>>
>> >>>
>> >>>
>> >>>_______________________________________________
>> >>>AusNOG mailing list
>> >>>AusNOG at lists.ausnog.net
>> >>>http://lists.ausnog.net/mailman/listinfo/ausnog
>> >>>
>> >>>
>> >>
>> >
>> >
>> >
>> >--
>> >
>> >Jared Hirst
>> >Chief Executive Officer
>> >Office Address: 2/2 Teamster Close, Tuggerah NSW 2259
>> >Postal Address: PO Box 3187, Tuggerah NSW 2259
>> >Phone: +61 2 8115 8801 | Network Ops: +61 2 8115 8850 | Main Switch: +61 2 8115 8888
>> >Web: www.serversaustralia.com.au
>> > How are we doing?
>> >please click here to tell Servers Australia - We‘re listening
>> >Notice: This message may contain private and confidential information intended only for the recipients. If you have received this message in error please delete immediately and notify the sender, as any distribution or reproduction of this message is prohibited. The views & opinions expressed in this e-mail are NOT necessarily those of Servers Australia
>> >Please consider the environment before printing this email - every year we are losing 40 million acres of oxygen producing forests through logging and land clearing.
>> >
>> >_______________________________________________
>> >AusNOG mailing list
>> >AusNOG at lists.ausnog.net
>> >http://lists.ausnog.net/mailman/listinfo/ausnog
>> >
>> >
>> >
> 
> 
> 
> -- 
> 
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> http://lists.ausnog.net/mailman/listinfo/ausnog
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20141022/7c223f21/attachment.html>


More information about the AusNOG mailing list