On Tue, Nov 11, 2014 at 01:18:02PM -0800, Tim Hoffman wrote:
All, Wea**ve had some interesting announcements of Megaport and IX Australia building internet exchanges into NZ over the last week or so. Ia**ve been a bit surprised at how little chatter there has been around this on the list. I thought Ia**d put some thoughts down (as someone who has worked for a bunch of networks in NZ and is now overseas working for big content) with hope of kicking off a discussion. At a 10,000 foot view, New Zealand has had a reasonably successful operation of IXs over the last 10-15 years. All of ya**all except for the two big operators openly peer. This is awesome! However, as has been discussed in other posts recently; 1/ More and more of the internet is becoming about connectivity from CDNs to users. 2/ More and more CDNs are coming to New Zealand and Australia
To my mind anything that brings about this happening more is good.
3/ Trans-tasman capacity isna**t hugely expensive
But for US content it's better if connecting directly to the US rather than bouncing via Australia.
Addressing points (1) and (2), as content comes into New Zealand, we have to remember that on a global scale, New Zealand is absolutely tiny. Ita**s a hard business case to get a POP there as a content provider. Content
Yeah.
providers that do come are likely just going to put in a single location - Auckland. So while WIX may make sense for ISPs in NZ peering with each other, for the bulk of the *interesting* content on the internet that you can get through peering, youa**re going to have to get it in Auckland.
Yes.
Networks have outages, planned works, etc. So assuming that most of NZa**s interesting content is in Auckland (with a bit more over the Tasman, which isna**t too expensive to get to these days), it makes sense to have multiple (redundantly built and managed) ways of getting at this content. Ita**s interesting looking at the differences between the 3 IXs wea**re about to have. * Citylink, who have the critical mass of all the peers, are in the most locations, and are the most expensive. Theya**ve also got a product set around metro ethernet, dark fibre, and have a CDN running. And theya**re doing some interesting things around a a**SDN driven IXa**.
Do content providers care about that part?
* Megaport, who have placed huge focus on their a**Virtual Cross Connecta** product and being able to configure this on the fly, but also have run a pretty successful IX.
To my mind - if we can get US providers to get a cross connect from California back to New Zealand running all their transit back to the US through this cross connect and have both their local peering and cross-connect for transit die at the same time that's a good thing. A while ago Valve/Steam kept dropping out on Highwinds who they connected to Telstra and Megaport with. If you had a forward route or a reverse route it would break for a significant duration. (slow BGP full table loads?) And if one of the paths went over Equinix that they directly connected with either having issues could create an issue. When there are assymetric paths and slow BGP failover times failover times can be much to long for interactive users.
* IX Australia, who are very IX focused, and have gone to great pains to ensure that thata**s their only product, making them the most a**independenta**. IX Australia also have public graphs of traffic levels through their IX (i.e. for NSW-iX http://monitor.nsw.ix.asn.au/cacti/graph_image.php?action=view&local_graph_id=91&rra_id=2) - which is commendable (it would be nice to see Megaport and Citylink follow suit on this one)
I dunno if it really helps that much. It's more important how much each network does over a peering exchange than the grand total.
I suspect youa**re going to find that most of the content will try to get to all 3. Ita**s my hope that most of the bigger ISPs try to get to all three, then the smaller ones can go to a couple, get most NZ routes
Well last I knew Valve still hadn't gotten onto Megaport directly. I would like to see Valve peering in New Zealand too myself. I think that once you say that someone should have multiple upstream transit providers and multiple peering links it makes the idea of getting onboard less attractive. And then you have to host gear remotely somewhere as well, and get cross connects. Having a simple redundant port setup, with onsite remote hands is a lot more attractive than having external contractors and gear somewhere like the sky tower.
reliantly over peering, and then transit the rest if it has to happen. It would be disappointing if we didna**t see the top 5-6 IXs + all the content providers on all of them. Also ita**ll be important to ensure that the result isna**t just that we have 3 split IX communities, all with different content/users on them without crossover, as this would cause a huge barrier to entry to smaller players if they have to connect to all 3.
From that point of view there's a few people on APE already - and having a second exchange would create redundancy. I think that redundancy is more important for local content than remote content. But it seems like it's pushing it to have 3 exchanges rather than 2.
Coresite having the split exchange between Los Angeles and San Jose is akin to having shared fabric between Wellington and Auckland, and although it simplifys and gets the number of peers up I think something less seemless would be better. In New Zealand all international goes out Auckland anyway, so it's normal to have transit from Wellington to Auckland, and less demand in the other direction.
For the community, having this competition is going to be a win. It wona**t kill APE, which has huge critical mass, and has done a great job in ita**s long history, but will ensure that all 3 IXs keep each other honest. What will be interesting to see is who turns up at which IX and how they all grow together.
Well what keeps popping into my head is that Telecom still don't peer, and that the US has had considerable peering issues recently. Any with the US recently changing daylight savings and California only being 3 hours ahead of us (well 21 behind) it pushes us closer to US peak time. It doesn't seem like the larger ISP's in the US want to improve quality, and instead want to charge for people to send them content. And I don't think that New Zealand is immune from these issues, and I wouldn't be surprised when more NZ content does turn up if we're going to run into more significant issues with basic Skype calls etc. If content providers did have peering in NZ, and a connect back to the US, then they'd still have to either pay Telecom to send to them, or to pay some other provider to send to Telecom, .. and if they pay some other provider, and that other provider oversubscribes (like Cogent did) then normal users doing normal peer to peer activity could be disrupted too. Which to my mind is worse than some user having buffering on something like Netflix etc. Ben.