Last week the Amazon AWS meetup was hosted at Optiver.
We had two items on the agenda: Eucalyptus and Optiver.
First off, Eucalyptus strategy of mirroring the Amazon API for their private cloud offering appears to be paying off, as Amazon and Eucalyptus recently announced their partnership.
It is good to see how this highlights the importance of a cloud API. Better yet, the essence of cloud computing is that you can automate everything - an API is not just a nice to have.
The big players in this area have shown that APIs, or standard interfaces, are essential. Automation requires code, and without standard interfaces, there is no automation, and no cloud computing.
The advantages of using the AWS api benefit both the customers, Amazon, and Eucalyptus. It allows organisations to create a web application hosted in the Amazon cloud, and migrate back and forth between public and private cloud as needed. As Eucalyptus had to admit, this mechanism was used with a customer to upgrade their private infrastructure to the newest code base. (And this new code base should now have the hooks in place to upgrade to future releases in place.)
An important difference between an AWS solution and a private one would be two things: In a way, the cost and the SLA of the AWS cloud are part of the API, and a private cloud API covers neither. Cost, Availability and Quality of Service are your own responsibility, and perhaps not as transparent. They may be orders of magnitude better, but will your private cloud provide the metrics to prove it?
Next was a presentation by Optiver's IT manager, who wondered what cloud computing had to offer their business. The company is a weird one in that sense, as their core business resolves around the 'money loop'.
The money loop is literally a very tight loop of memory, cpu and network between the stock exchange and the trading algorithm. The algorithm may be smart, but the most important thing is that it must be fast. In online trading, timing is everything, and traders like Optiver reduce delay by running on high speed networks, a minimal number of hops in the network.
Network speeds are about capacity, latency and variation. Bits, bytes, and packets add up. Network capacity is not relevant here, but the time a packet needs to get onto and off the wire definitely does. When I had to worry about latency, it was the mid-90s - we were happy if we had T1 and E1 access to Frame Relay (Network as a Service). These guys talk about Gbit and 10 Gbit speeds. And they don't even use a fraction of that capacity.
In short, IT Operations at Optiver is Business Operations. The teams need to work very closely - and they physically sit close to each other. When important issues come up, the people will find each other and sort things out.
Is there room for the cloud at companies like Optiver? As always - it depends. The company has offices, and needs desktops with normal office applications, none of which are immediately critical to the business. Office and trading infrastructure need to be separate anyway, so why not outsource office apps and email to a SaaS provider?
In the trading industry there is some talk about Algo as a Service - that would be a ultimate deal for a trader - to have their algorithms actually hosted at the exchange. But they would lose control over a lot of parameters - network speed/utilisation, in short Quality of Service.
I would expect an AaaS to have some very interesting SLAs and APIs.