Monday, August 1, 2011

Queries Per Second - RTB Complexities

Ramsey McGrory of the Right Media Exchange recently said in an AdExchanger article:

"For example, it's very difficult for a DSP or an agency trading desk to consistently respond to 35,000 QPS (queries per second) no matter the stated SLA (service level agreement) because their infrastructure may not be able to handle that load or they may not have enough data to have a point-of-view on all of the different inventory."

As someone who is tasked with sorting out this issue for a digital media technology supplier this is a VERY real issue.    Sorting out how to do this is not a simple thing.   There are two main issues that make this horribly complex:  Speed and Data Scale.   


So lets take each one seperately and talk about what it means.   The first one "Speed" is quite simple.    Dealing with 35,000 queries per second is quite the computational problem.   You need to evaluate a massively large amount of queries and do so in a way that you respond within 15ms.   The trick or challenge in this whole thing is make sure that you handle the number of queries in a way that deliver upon your brands value proposition to the market effectively.   I can't release how we are cutting this but I can say it required me to pull in resources from wall street and silicon valley.

The next is data scale.    Data scale is a real tricky one.    If you cookie the consumers or base it on profiles you have a pretty implicitly scale issue because you are required to see the use again and the data is prone to issues of decay on the cookie variables.  The trick is to find a slice of data that the brands find as premium for targeting ad environments that supports the brands goals.

All of this is "tres" complicated but none the less a fun problem for a nerd like me.   

I
LOVE
THIS
SHTUFF!     

The main problem I see in this whole equation is that ad agencies are going to struggle to define their position in all of this because they are not primary data owners.   At best the data they have represents aggregates of other peoples data with post click data by chance.  This has a few problems to it...  

1) Aggregate matches take time and massive infrastructure.  Technology infrastructure is not in the agencies DNA.    More so the aggregate game of multiple cookie matches gives nothing more then DR optimization levers.  The need to create their own down past the aggregate becomes absolutely needed.

2) As $$$ continue to flow into the ecosystem the cost of everything rises as its a demand driven market.  The value of the middle men becomes tougher and tougher to justify as the squeeze begins to really hit as demand raises the supply price.  The squeeze means its difficult to justify high technology costs required to invest in problem #1 infrastructure and data.

2 comments:

  1. Hey sweet cheeks, are RTB exchanges vulnerable to manipulation like HFT or is this apples & oranges?

    http://www.nanex.net/20100506/FlashCrashAnalysis_Part4-1.html

    While analyzing HFT (High Frequency Trading) quote counts, we were shocked to find cases where one exchange was sending an extremely high number of quotes for one stock in a single second: as high as 5,000 quotes in 1 second! In many of the cases, the bid/offer is well outside the National Best Bid/Offer (NBBO).

    Volume of these erroneous quotes can be enormous on any given burst or sequence and can last anywhere from a few hundred milliseconds to a few entire seconds.

    If each reporting exchange for each stock quoted at 5,000 per second this would work out to 180,000,000 quotes per second.

    Our recommendation for a simple 50ms quote expiration rule would eliminate quote-stuffing and level the playing field without impacting legitimate trading.

    ReplyDelete
  2. This specific issue isn't replicable in principle within these exchange frameworks and I am happy to go into why if you see value in that but the point you bring up is a real one. Technology can now impact market price and activity. I have been thinking about and how people could "game" the system and I am in agreement that this is something the media industry must be very careful with in the coming months. Happy to discuss some of these thoughts and principles but prefer it be offline and not on comment thread.

    ReplyDelete