Berkeley CIFT 

Back in the old days, when stocks traded, slowly, in 1/8s and in one place, it was easy to measure the state of a market. Look at the recent quotes and trades on a ticker or display.

Now, with penny pricing, dozens of market fragments, limit orders gone in milliseconds, and dark liquidity the only way to infer the state of a market is to do a great deal of math, in a hurry. Some fragments have LOBs, some use other market structures.

A simple LOB real time analytic is price to fill a given size order. But when a large portion of liquidity is dark, it doesn't show in the LOB, but in higher refresh rates as programmatically hidden orders appear.

There is a high level of structural parallelism in marktet analytics - most of them deal with only one security at a time. In the US stock market, with roughly 7,000 tradeable names, thousands of specialized processors could be used simultaneously.

Our EECS Parlab colleagues asked us to find some actual working parallel market analytics to get things rolling on a real application.

Take a look at their site http://parlab.eecs.berkeley.edu/
A major theme is the need to drive research with real applications. There are many lessons in parallel computing about avoiding solutions in search of problems.

We have put together a collaboration in this area, and hope that conversation here will be part of it.