07.02.2013 Views

Best Practices for SAP BI using DB2 9 for z/OS - IBM Redbooks

Best Practices for SAP BI using DB2 9 for z/OS - IBM Redbooks

Best Practices for SAP BI using DB2 9 for z/OS - IBM Redbooks

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

(max_size=250,000,000, and max_split_parts=40).<br />

When the data arrives at the blade, the RFC compressed data is uncompressed.<br />

Then gzip compression is done be<strong>for</strong>e writing temporary objects to GPFS, in the<br />

<strong>for</strong>m of *.tmpx files. Each blade has its own unique storage location/<br />

subdirectories <strong>for</strong> the same cube. The next phase is to read these temporary<br />

objects that were just written to GPFS. Initial indexing and optimizing take place.<br />

This uncompresses the data, reorganizes it, and compresses again with a<br />

different algorithm. Lastly, the data is written as a whole to the GPFS as final<br />

persistent objects in the <strong>for</strong>m of *.bin files.<br />

In Appendix D, “Project Jupiter: evaluation details” on page 325, Examples D-6<br />

through D-8 display a list of <strong>BI</strong> Accelerator indexed cubes, a cube example with<br />

40 part horizontal partitioning, a list of all columns of an indexed cube, and their<br />

corresponding temporary and persistent objects. They illustrate <strong>SAP</strong> <strong>BI</strong><br />

Accelerator horizontal partitioning, vertical decomposition, and compression.<br />

The <strong>BI</strong> Accelerator index creation process is very I/O intensive to the GPFS<br />

storage shared by the blades. During the test, we started with the default value of<br />

256 K <strong>for</strong> the GPFS I/O block size. We added a second DS8300 storage system<br />

to handle the high GPFS I/O activity. Our further investigation found that the high<br />

GPFS I/O activity was actually due to the large block size of 256 K, so we<br />

experimented and determined that blocksize=64K and maxblocksize=64K<br />

yielded better GPFS per<strong>for</strong>mance.<br />

The two DS8300 storage systems were configured <strong>using</strong> RAID-10 because of<br />

the write intensity of the <strong>BI</strong> Accelerator index creation. In general, RAID-10 is<br />

recommended <strong>for</strong> write-intensive workloads. However, it requires more disk<br />

space.<br />

Chapter 11. Project Jupiter: large <strong>BI</strong> Accelerator scalability evaluation 243

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!