IBM Systems Journal - 2002 Copyright

IBM Skip to main content
  Home     Products & services     Support & downloads     My account  

  Select a country  
Journals Home  
  Systems Journal  
    Current Issue  
    Recent Issues  
    Papers in Progress  
    Author's Guide  
Journal of Research
and Development
  Contact Us  
  Related links:  
     IBM Research  

IBM Journal of Research and Development  
Volume 42, Number 2, Page 347 (2003)
Storage Systems
  Full article: arrowHTML arrowPDF   arrowCopyright info


Characteristics of I/O traffic in personal computer and server workloads

by W. W. Hsu, A. J. Smith
Understanding the characteristics of I/O traffic is increasingly important as the performance gap between the processor and disk-based storage continues to widen. Moreover, recent advances in technology, coupled with market demands, have led to new and exciting developments in storage systems, particularly network storage, storage utilities, and intelligent self-optimizing storage. In this paper, we empirically examine the physical I/O traffic of a wide range of server and personal computer (PC) workloads, focusing on how these workloads will be affected by the recent developments in storage systems. As part of our analysis, we compare our results with historical data and re-examine some rules of thumb (e.g., one bit of I/O per second for each instruction per second of processing power) that have been widely used for designing computer systems. We find that the I/O traffic is bursty and appears to exhibit self-similar characteristics. Our analysis also indicates that there is little cross-correlation between traffic volumes of server workloads, which suggests that aggregating these workloads will likely help to smooth out the traffic and enable more efficient utilization of resources. We discover that there is significant potential for harnessing "free" system resources to perform background tasks such as optimization of disk block layout. In general, we observe that the characteristics of the I/O traffic are relatively insensitive to the extent of upstream caching, and thus our results still apply, on a qualitative level, when the upstream cache is increased in size.
Related Subjects: