I have a lot of traffic to deliver to China. How is this best done?
Funny you should ask. At AfPIF, EPF, NANOG and CHINOG over the last month, I discovered two interesting things about interconnection in China.
First, the dominant ISPs in China (China Telecom and China Unicom) have each expanded across both the north and the south now. The country used to be roughly divided into China Telecom in the north and China Unicom in the south. But even with this expansion of market coverage, they peer with each other in only three locations, with multiple 100G interconnects in each location, and each of these interconnects is congested.
The Great China Sieve
“Why congested? Why wouldn’t they interconnect with more capacity?” I asked.
Well, it turns out that keeping the interconnects congested forces customers to have to buy transit from both China Telecom and China Unicom. My dad always said to be careful when assigning motivations, but this motivation came from people very close to the situation there.
This is not new. As Mark Twain once said, “history may not repeat itself, but it does rhyme.” This tactic of selling more transit by letting interconnects congest has precedent in other places in time and place as mentioned in my book and in a previous blog.
So the best delivery option today requires buying transit from both China Telecom and China Unicom to avoid your packets traveling across the Great China
Then the question become, where to buy access to CT and CU routes?
Where to buy China Network Routes
The second thing I discovered is that deploying network into China is a prohibitively expensive and difficult proposition. Many have tried and found that giving up of 51% of control to a local China business to be problematic. Intellectual property and operational expertise are in some cases considered the business’s crown jewels. Others have shared their long lists of deployment and operational snags, including import duties on equipment, licensing delays, equipment being held up in customs for no reason, corruption, difficulty getting information or cooperation from others, etc.. In general I heard deployment stories into China that were more fatalistic than most international deployment stories. But some have persevered and, because it is so difficult, they have something now to offer the world - the ability to not have to build in to China themselves !
So building into China may be off the table, so “Where does one buy China Telecom and China Unicom routes outside of China and how much will it cost?”
The Market Prices for China Routes
It may sound counter intuitive, but the closer you get to China, the more expensive it is to get direct routes into China!
Some rough price points collected from the field last month demonstrate this.
If you buy transit in China you might pay $100/Mbps.
Buy China Unicom transit in Hong Kong, right next to China, and the price is about $75 per Mbps.
If you instead buy China Unicom routes from London, the price is around $5/Mbps.
In Los Angeles, while free peering might not be had, the price for transit to access routes from China Unicom and China Telecom might come in towards $1 per Mbps.
The pricing is approximate and varies widely but the trend is pretty clear: the farther away from China, the cheaper the price.
My recommendation. I like the idea of buying China routes from both China Telecom and China Unicom in Los Angeles, and even paying the cost for tethering (aka “remote peering”) to get into Los Angeles to do so.
(Full disclosure - I work for IIX, a company that sells peering solutions including tethering.)
The reason I recommend tethering into Los Angeles (if you are not already there) is that I would rather trade latency for better direct connectivity to China any day of the week. Packet loss is the #1 killer of performance, with latency and jitter coming in as a distant thirds, depending on the application. Packet loss causes the TCP window to divide by two, slowing the transfer rate to only half of the available data transfer rate in the event of packet loss. Buying access to China by tethering into Los Angeles get you get direct (albeit with slightly higher latency) access to the dominant China players for the best metered price, and not having to have a colocated presence makes the business case even stronger.
A transition then to a physical presence in Los Angeles can remove some of this latency to China. That is, assuming the content for China is located or cached in Los Angeles. If you are already in Los Angeles then you are one step ahead of the game and the business case for buying transit to China there is stronger still.
The optimal blend of interconnect and traffic delivery solutions is really dependent on the application network requirements of course, but the congestion context I learned about and these price points both provide incentives to tether into China rather than to build towards China.
I hope this helps -
P.S. If you have China deployment and operations stories that you can share with the others on this blog, please send them to drpeering at drpeering.net