zur Startseite


Analysis and optimization of storage I/O in distributed and massive parallel high performance systems  (tbd)
Bearbeiter Salem El Sayed
Betreuer Dipl.-Inf. Simeon Wahl
Prüfer Prof. Dr.-Ing. Sven Simon

Kooperation mit IBM Böblingen (confidential)

Although Moore's law ensures the increase in computational power, I/O performance appears to be left behind. This minimizes the benefits gained from increased computational power, since the processors have to idle for long times waiting for I/O. Another factor that slows the I/O communication is the increased parallelism required in today's computations. Most modern processing units are built from multiple weak cores. Since I/O has a low parallelism the weak cores will decrease the I/O performance. Therefore by analyzing the I/O stack under the new conditions of multi-core and massive parallelism, some conclusions can be drawn as to how such I/O access can be improved.
The following optimization are to be taken in this thesis:

  • Optimizations on the I/O stack to accommodate parallelism and increase performance.
  • Optimizations of the disk access patterns and file-systems for introducing Solid State Drives (SSDs).
  • Use of idle cores during an I/O transaction to perform on the fly I/O data compression.

These optimizations on the lower level of I/O could then be projected on high performance computing systems with massive parallelism to achieve better total I/O access time.