LOGIN
>> Home
>> Topics
>> Students
>> Partners
>> Statistics
Information for topics
Topic Id:
ID topic:
238
Partner Email:
srba@cs.aau.dk
Project Title:
Spatial Data Management on Multi-Core Processors
Abstract:
For many years, advances in mainstream computing hardware were mainly concentrated in increasing the clock speed and capabilities of single-core processors. Recently, this approach seems to have reached its limits and the current trend is to increase the number of cores on a single CPU chip. This development is likely to continue in the future and this means that main-stream computers will be capable of running large numbers of truly parallel threads. Using such parallelism to the greatest extent possible is an interesting challenge. Another hardware development is the constant increase in main-memory sizes. Modern mainstream computers come with large amounts of memory, but the access to this memory is relatively slow, when compared to CPU speeds. To solve this problem a memory hierarchy of CPU caches is employed. Thus, parallel computing environment can be described as multiple parallel threads accessing a large amount of main memory through a shared hierarchy of caches. Note that this is quite different from parallelism that is possible, for example, in the environment with multiple CPUs each with a modest amount of main memory. From the applications side, spatial data management is a rapidly growing discipline where large amounts of data are processed with computation-intensive algorithms. In a spatial database, parallelism can be used in two ways. First, expensive operations, such as spatial joins can be parallelized to speed them up. Next, queries and updates from multiple users can be processed in parallel. The goal of the project is to explore how parallelism of multi-core processors can be best exploited for management of spatial data in main memory. The project is quite open. A number of different questions could be investigated: How existing abstract models for parallel programing (for example, Map-Reduce) can be used to do spatial data processing? Such models may allow easy programming, but may not result in the best utilization of the available parallelism potential. What kind of concurrency model is best? To what extent advanced hardware features, such as atomic operations can be used? In contrast to more traditional spatial applications, new kinds of applications may receive massive amounts of data updates (for example from censors) which must be processed concurrently with queries. A possible direction of the project would be to explore the use of GPUs (graphical processing units). GPUs, available in most modern computers, have even more cores and a more complex memory hierarchy. Independently of the direction taken, the project would involve a lot of experimentation.
Advisor:
Jiri Srba
Link:
Degree:
Master
Keywords:
