Yeah yeah, a few of my friends who work in the financial markets told me the same.

We were saying mostly in jest… Losing a DNS packet stinks, but not as much as a trade, I’m sure.

Let me explain.

Each NYSE order has a few data params - symbol, price, volume, whole_lot? (do we only execute if we can trade all of the specified volume), party_id, etc. When you send an order, it gets put in the “book” for that symbol. We’ll ignore orders that say “buy/sell at the current price” - those are uninteresting.

The “book” is an array of fifos. Elems in the array are keyed off of the price. Elems in the fifo are based on time of submission and order type (buy vs sell).

Then, a timer goes off, and the trading engine matches orders that match on price (and sometimes size) based on the last trade price, whole_lot?, time of submission, etc. This is a logical puzzle for those who care.

The easy part of the NYSE’s system is that it’s easy to parallelize - one machine per book - but also quite a logical mess. Add to that, the need for audit logs, and well, it’s a much bigger job to handle 1 NYSE transaction vs 1 DNS transaction.

In other words, the NYSE’s volume is a little more complex and much more computationally complex.

Not to say the OpenDNS is not impressive, but it’s a different type of problem. =)

jb

Well done guys, i hope it now or in the future equates into big profits for you company.
S
tay well

I doubt the size of each individual request is very different. I know DNS requests are small. Need to find someone who knows more about the nature of the NYSE messages to know for sure.