The IP Development Network Blog

Back in April, when I was writing my original "Joost: analysis of a bandwidth hog" piece, I had no idea that the piece was going to make up 24% of my pageviews since then. It is quite possible that this had nothing to do with my writing and everything to do with the strategy secrets to be found in the link colmmac posted in his comment. Unfortunately, these secrets are now gone but you can read all about it here.

Whichever it was, it was clear that people were interested in how the application was behaving, so I updated the analysis in late May with my "Joost about on track" post that also looked at the wider strategic issues facing the company. My summary was that they were kitted up to play, but by no means certain of making the cut. They had (and still have) a lot of work to do on their EPG to improve what is a very static and "been there, done that" channel guide.

The content delivery network though was improving and this has clearly moved the company towards the concept of being "a high quality ad-supported secure cost-effective delivery platform" that works within networks rather than over the top. Their new CEO used to head Cisco's network equipment division, remember... Joost: the operating system for IPTV?

Now in mid-July, having run the analysis for a third time, it is clear that one thing that is working really well for them is the peer-to-peer network. Success here is measured in hard currency: the bill they pay their network providers (Level3 mainly, but also BT Infonet). What I am looking for in my analysis is the proportion of traffic that comes from peers versus the proportion that is served from these centralised locations.

In order to get my results, I am using Wireshark, which many readers will know as Ethereal. This allows me to see all the traffic on my LAN, and if I turn everything off except Joost, enables me to see what the application does when it is active.

What this shows is that for popular content at least, Joost has successfully offloaded the vast majority of traffic from its paid-for connections onto its free P2P network (free to Joost that is). In April's study 47% of bytes came from central servers, six weeks or so later on the 25th May I reported that this was at 18%. Results from 11th July show that this is now 6.7%

This popular content (the Fifth Gear Ferrari 430 Spider road test) is now well seeded, such that no one peer delivers more than 12% of the total file. There are some major sources: 6 peers between them give me 50% of what I am after and 14 peers account for 80%, but there is a long tail.

In total, there 99 IP addresses on 75 different networks in 33 countries delivering traffic to me.

From what I have seen, the network is now starting to prefer peers on this side of the Atlantic (possibly because there are just more of them now). Where in my May sample there was 39% coming from the US and Canada, this is now 4% in my July data. This is good news. Things are moving in the right direction.

So now that Joost's distribution capacity is seeded in Europe, expect to see this get more country specific to align with the network interconnection arrangements already in place. Serving Joost punters in the UK from other users in the UK means that ISPs costs are mitigated as they will likely avoid the most expensive routes - Transit.

But this is still only mitigation. The overall burden on the ISPs goes up with every Joost user that leaves their PC on even if the application minimised.

As I have mentioned before, Joost also uses the network when it is inactive (minimised), which I have questioned in the past. I don't know if this is new or if I missed it the first time, but the Terms and Conditions acceptance box now states clearly "I know that the Joost software will operate when minimised unless I fully exit the software".

Perhaps this is a necessary evil, but in a world where many people either have bandwidth limits or fair use policies set by their service providers, consuming someone's scarce resources for use by a complete stranger is still a big concern for me. This is where the house of cards could be vulnerable - you might give them a lift if you were going that way anyway, but would you let a stranger borrow your car when you are not using it (even if you were 100% sure you'd get it back)? Hmm...

Perhaps this won't be an issue. Certainly when I was sampling yesterday, I was surprised to find that the Joost application was not using my bandwidth when idle. Granted, this was yesterday morning around 11am so I can understand that demand from other users will be pretty low at that time - most people are at work of course and not watching Fifth Gear on Joost. I am certainly going to have to come back to this point and look again what is happening at peak hours.

I speculated in June that once the peers are well established, the file sizes can be cranked up to increase resolution and user experience. That time may well be approaching - perhaps when the service finally goes from the most well known Beta trial in history to being the real thing?

Having said that, to my untrained eye, the resolution already looks much better than it did in April. This might well be new codecs in the ever evolving software (the latest download is version 8 of version 0.10, so the guys there have clearly been busy). Whatever it is, the picture quality is now almost as good as standard definition TV. Not quite up there with Channel4's on Demand service (which uses a lot more bandwidth), but a lot better than Babelgum.

It does not look like this improvement has come by cranking up the bandwidth used by the application. My traces show that this is about where it was before, if anything the use may even be ever so slightly lower. The picture quality though is a lot better.

So that was popular content, what about the stuff that is buried in the Joost channel guide? Wow, do you have to work to find something watchable here? In fact I didn't watch it, I put it on, started Wireshark and went to make a cup of tea.

I chose the Community Channel because it seemed like the sort of thing that trendy Joost-heads would be most unlikely to be interested in. To be fair, the film was actually very well made, home made, but well made. It was about a couple of schools who decided to give their pupils a taste of the business world by letting them play Tycoon where they could design and develop a product for sale in their communities. Thankfully Peter Jones was nowhere to be seen.

So I was surpirsed when Wireshark revealed that this show about three very British schools, was coming largely from servers in the US. 68% of bytes came all the way across the Atlantic while almost all the rest was from the UK. The source networks? Level3 in both cases.

The aim of the experiment was to see how long tail content was distributed and here we have an answer. It makes me wonder (again) about the whole Long Tail thing. It is going to cost Joost an awful lot more to distribute this as it will be very hard to seed. They will be much better off if / when they get mainstream content on there that they can deliver as timeshifted linear TV.

My research has previously shown that on YouTube at least, the internet means that the richer get richer. The 20th most popular clip there had 18% of the views of the most popular, while the 20th most popular DVD rental generated 79% of the revenue of the no. 1 title. Humans like to follow the herd and social networking helps by showing the direction the herd is heading.

If it is far more expensive to distribute the unpopular stuff and demand for it is much lower (lower demand = lower prices), where's the business case? Time will tell, and I've been wrong before...

Labels: Joost, timeshift TV, video on demand

# posted by Jeremy Penston @ 7/12/2007 03:21:00 PM

Thursday, 12 July 2007

Joost: further analysis of a bandwidth hog