+44 (0)1926 623303

Knowledge

11/02/2015

What is DB2 BLU Acceleration and how does it work?

IBM DB2 BLU Acceleration is a collection of technologies that are seamlessly integrated in DB2 10.5 and above, and can be used by tools such as IBM Cognos and IBM SPSS to provide ‘blazing fast performance for analytical workloads’. With queries being answered 20 or many more times faster, BLU enables instant insight using less processing resource.

How does BLU Acceleration achieve these speeds?

BLU Acceleration incorporates a range of innovations including column organised storage, actionable compression, in-memory performance, data skipping and CPU acceleration. Whilst DB2 on it’s own has features to optimise data retrieval, BLU Acceleration takes this to a whole new level.

Columnar organisation for easier and faster analytics

BLU Acceleration enables data in DB2 to be stored in column organised tables rather than row organised tables. BLU maintains the row relationships but when storing data on pages it does so by column. This improves storage compression efficiency and makes it easier to use through simpler design and tuning. Most importantly it also means much improved analytics query performance when summarising data. Some OLAP queries, depending on individual workloads, configurations and conditions, are processed over 100x faster than a similar row based query.

DB2 BLU Acceleration also features a columnar pre-fetching algorithm to work out which data to send to RAM, just a few milliseconds before it’s loaded. Therefore reducing the data required in RAM to process a particular query.

Actionable compression to efficiently compress and store data

DB2 already has algorithms that optimise data retrieval, such as off-set coding. BLU Acceleration enhances this with order-preserving frequency compression and page level compression efficiency of having many values for the same column on one page. If a number is frequently retrieved BLU Acceleration stores this value at a higher level than those numbers that appear less often. So, data can be compared in a compressed state, avoiding the heavy workload of decompression, reducing data on disk and in memory.

In-memory enhancements to optimise performance

BLU Acceleration introduces more than system memory optimisation; it introduces in-CPU memory optimisation. Every aspect of BLU Acceleration has been designed to minimise access to RAM and maximize processing time in L3 and L2 caches, which operate at an order of magnitude faster than RAM.

Compared to DB2, BLU Acceleration uses more in-memory processing techniques to optimise access to RAM as and when required. If the data size grows beyond the RAM capacity it will automatically store the intermediate results to disk.

Memory access is also optimised within BLU Acceleration because the majority of access is via a CPU cache rather than RAM. Therefore latency is minimized and the CPU is kept busy, achieving further CPU performance improvements.

Data skipping for faster queries with no fine-tuning

Data skipping enables BLU Acceleration to save CPU, RAM and I/O resource by skipping large sections of data that do not qualify for a query. The user does not need to take any steps to tune this. BLU automatically records metadata about each c.1,000 records of data so that when a process is underway the metadata is checked to establish if there a value in that range which is needed. If not then the entire range is ignored.

Unlike the Zone Map functionality of the Pure Data System for Analytics this metadata is against the ‘range’ of data, not tied to a particular page or extent boundary. This means BLU Acceleration can find the ranges needed via metadata, ignoring all other data sets to significantly reduce processing time.

CPU Acceleration via SMID

Where available BLU Acceleration uses low level CPU instructions in the single instruction, multiple data processing (SMID) hardware feature to process multiple points of data simultaneously and/or retrieve results from multiple data elements from the same register. If the SMID feature isn’t available on the hardware then BLU Acceleration will emulate the SMID capability using software to deliver some of the benefits of SMID.

BLU Acceleration provides what IBM describe as ‘breakthrough performance’. It is simple to use, dramatically fast and relatively easy to set up as ‘load and go’.

EBI are experts in DB2 and an IBM Premier Business Partner. If you want answers about BLU Acceleration please contact us.

 

 

IBM DB2 BLU Acceleration video – live demo

Comments