eXtremeDB delivers on-chip analytics, for breakthrough efficiency working with market data. On-chip analytics is implemented in software, and therefore should not be confused with the practice of optimizing analytics using FPGAs.
On-chip analytics also differs substantially different from in-memory computing. While in-memory computing uses main memory as DBMS storage, to eliminate various kinds of latency, on-chip analytics optimizes data flows to and from the CPU and maximizes the amount of relevant data held in CPU cache, to reduce latency-inducing transfers between CPU cache and main memory.
eXtremeDB on-chip analytics can be used independent of the product’s in-memory database system (IMDS) capability.
On-chip analytics in eXtremeDB has three components:
- The “sequence” data type implements columnar layout for data elements. Sequences are ideal for representing time series such as tick streams.
- A rich library of vector-based statistical functions, to accelerate analysis of these sequences/columns.
- A pipelining technique is used to combine these functions into assembly lines of processing, with output of one function becoming input for the next.
Pipelining keeps market data in CPU cache while it is being processed by multiple functions, slashing latency by eliminating transfers between the CPU and main memory.