Saturday, 14 September 2013

MPP(Massively parallel processing) & SMP

Hi Guys,

We come across terms like MPP (Massively parallel processing) while going through documentation for teradat, datastage and other data processing or ETL tools

MPP is a general term which can be applied for clustur of computers where on one server rack you have multiple cpu usually in 1U casing. also multiple racks are connected together and you have a clustor

It can also be called for grid computing where commodity hardware( normal desktop pc) are connected for computing .

SMP (Symmetric multiprocessin)

In this 2 processors share the same memory (RAM) in olden day they had two processors on same motherboard sharing same RAM .Now all laptop above dual core are actually SMP . they have 2 in build processors and sharing same memory. SMP is great for multithreading and multiprocessing.Processes are broken into threads and supplied to individual processors

So what is a process 

A computer program is passive set of instructions.Process is the actual execution of those instruction with its current state being saved in memory.Usually if you open a program twice you will see two processes in task manager

So whenever you open a program it occupies part of RAM memory.Threads use a part of this memory allocated to process. then in case of dual cores this threads are then given to two different cores

Multithreading makes it possible to utilize cpu to max if the thread running does not utilize full cpu other thread can take advantage.Similarly if two threads are working on same resulset cache can be shared.In case of single cpu time sharing can be done among threads

So How much cpu utilization is normal.

If you open task manager you will see that cpu is utilized less than 30% why is that.dont we want our CPU to be utilized to 100%


Very few data bases are MPP (Shared nothing architecture) based .

Oracle 11g is shared everything SMP

Oracle Exadata is nearest to MPP (Not sure if fully MPP)

Terada is MPP

HP Vertiga is MPP

Kognito is MPP


No comments:

Post a Comment