Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Microbenchmark-based Extraction of Local and Global Disk Characteristics

Nisha Talagala, Remzi H. Arpaci-Dusseau and D. Patterson

EECS Department
University of California, Berkeley
Technical Report No. UCB/CSD-99-1063
1999

http://www.eecs.berkeley.edu/Pubs/TechRpts/1999/CSD-99-1063.pdf

Obtaining timely and accurate information about the low-level characteristics of disk drives presents a problem for system design and implementation alike. This paper presents a collection of three disk microbenchmarks which combine to empirically extract a relevant subset of disk geometry and performance parameters in an efficient and accurate manner, without requiring a priori information of the drive being measured. Novel among the benchmarks is the utilization of linearly increased stride to glean a spectrum of low-level details including head-switch and cylinder-switch times while factoring out rotational effects. A bandwidth benchmark extracts the zone profile of disks, revealing that the previously preferred linear model of zone bandwidth is less accurate than a quadratic model. A seek profile is also generated, completing a trio of benchmarks. Data is collected from a broad class of modern disks, including five SCSI, two IDE, and two simulated drives.


BibTeX citation:

@techreport{Talagala:CSD-99-1063,
    Author = {Talagala, Nisha and Arpaci-Dusseau, Remzi H. and Patterson, D.},
    Title = {Microbenchmark-based Extraction of Local and Global Disk Characteristics},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {1999},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/1999/6275.html},
    Number = {UCB/CSD-99-1063},
    Abstract = {Obtaining timely and accurate information about the low-level characteristics of disk drives presents a problem for system design and implementation alike. This paper presents a collection of three disk microbenchmarks which combine to empirically extract a relevant subset of disk geometry and performance parameters in an efficient and accurate manner, without requiring a priori information of the drive being measured. Novel among the benchmarks is the utilization of linearly increased stride to glean a spectrum of low-level details including head-switch and cylinder-switch times while factoring out rotational effects. A bandwidth benchmark extracts the zone profile of disks, revealing that the previously preferred linear model of zone bandwidth is less accurate than a quadratic model. A seek profile is also generated, completing a trio of benchmarks. Data is collected from a broad class of modern disks, including five SCSI, two IDE, and two simulated drives.}
}

EndNote citation:

%0 Report
%A Talagala, Nisha
%A Arpaci-Dusseau, Remzi H.
%A Patterson, D.
%T Microbenchmark-based Extraction of Local and Global Disk Characteristics
%I EECS Department, University of California, Berkeley
%D 1999
%@ UCB/CSD-99-1063
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/1999/6275.html
%F Talagala:CSD-99-1063