Chapter 24 Flashcards
How can we improve speed
- Hardware architecture
- Software architecture
- Indexing techniques
- Join techniques
What is VLDB
very large databases
What is parallelism
Divide the data set into pieces and assign to all processors equally. Then the processing has to done independently parallel on all data sets simultaneously
Does divide and conquer favors parallelism
Yes
What are the requirements for parallelize
All following 4 things
- Parallel architecture (e.g. symmetric multi-processors (SMP), clusters, massively parallel systems (MPP), non-uniform memory access (NUMA) )
- Sufficient I/O bandwidth
- Underutilized CPU
- Sufficiently memory
Is it possible that system performance degrade in parallelism
Yes
What are the factors for system speed
- Index usage
- Number of concurrent users
- Complexity of technique
- Amount of detailed data
- Complexity of data model
- Underline schema
- Size/scalability
What are the 2 methods to upgrade underlying hardware
- Speed up
- Scale up
What is speed up
More resources means proportionally less time for given amount of data
Does there be linear relationship between degree of parallelism and transaction/sec ideally
Yes
What is scale up
If resources increased in proportion to increase in data size, time is constant
What is the formula of speed up
Ts/Tm
How we quantify speed up in real time
Amdahl’s Law
What is parallelism in OLTP
How much transactions can be executed per second
What is parallelism in DSS
How big one complex query divides and executed to get result faster
What is parallelism in software architecture
- Shared memory
- Shared disk
What are other 2 categories of parallelism
Data parallelism
Spatial parallelism
What is MPP
In massively parallel systems, every node has independent processor, independent I/O and independent memory. All nodes are connected with interconnection network.
Does SMP supports multi tasking and multi processing
Yes
What are distributed shared memory machines
Its a combination of both architectures. One software architecture is shared disk RDBMS architecture
What is shared disk RDBMS architecture
Database tables are shared. There is a central node and processors access through it to desired section of data.
What is memory lock
In RDBMS architecture, memory lock is used to avoid more than one processors trying to access same table/data set.
What is disadvantage of memory lock
Process become serialized
What is software shared nothing RDBMS architecture
Every processor has its own independent memory. There is no memory locking over here. Once a processor is failed its all memory is gone.
Does shared disk RDBMS and shared nothing RDBMS architecture are logical architecture
Yes
