stock market level 2 orderbook algorithm

I need a more efficient algorithm that I do have in a current application. The algorithm should calculate a indicator based on a "level 2 orderbook". A level 2 orderbook is a list of all buy orders("bid") for different prices with different quantities; and also a list of all sell orders ("ask" for different prices with different quantities. Example: Ask 1.45 2000 Ask 1.44 1000 Ask 1.43 500 Ask 1.42 50 Bid 1.41 1000 Bid 1.40 10 Bid 1.39 500 Bid 1.3805 100 The level 2 book is calculated for each instrument. For a symbol it could contain 500 Bids and Asks. Whenever there is a new bid a message is received, say "Bid 1.41 1000". When a bid is removed from the list, a message with 0 quantity is received, say "Bid 1.41 0"; in this case the bid is removed from the orderbook. Same goes for the asks. Now the indicator is calculating the sum of bid quantities and also the sum of all ask quantities. Then the net difference is calculated as net difference = sum of ask - sum of bid. The percentage then is net difference / (sum of ask + sum of bid) * 100. This is the indicator we are interested in. Right now this indicator is recalculated from scratch for each message that comes in, whihc means we iterate through an array of possibly 500 items, and this costs time. The algorithm should be improved, so that the total bid sum, and the total aks sum is increased or decreased; therefore the 500 iterations can be saved. Because there can be 50.000 update messages a second, this means a significant performance improvement. ## Deliverables We also need a parameter, which would limit the calculation of the book only to the best bids or asks. In our example the best bid is 1.41 and the best ask is 1.42. If the range would be 0.02 then we would only use the ask till 1.44 and only those bids down to 1.39. With this calculation, the running sum has to be calculated on the "top bids" and "top asks", so it is a little more complex; but still should be much faster than a full recalc with each message. We have a complete visual studio solution with the inefficient algorithm and require that it is speeded up. Project is in C# and the programmer should be interested in mathematics and financial markets. We do have lots of follow up work. I would share a dropbox folder for those who are interested which contains the full project and also demo files to allow to simulate past market data. Because the algorithm is incremental, one wrongly added calculation would mean that the full history thereafter is wrong as well. Therefore to verify that the algorithm works, for debugging reaons I want to be able to verify this. So in this case, the full "non incremental" algorithm result needs to be compared with the incremental algorithm. This is to make sure the algo is correct.