home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Power Tools 1993 November - Disc 2
/
Power Tools Plus (Disc 2 of 2)(November 1993)(HP).iso
/
hotlines
/
wsyhl
/
compson
/
script1.txt
< prev
Wrap
Text File
|
1993-01-07
|
2KB
|
45 lines
Title: Disk performance & how to compete against Sun's IPI disks?
Workstation performance is a function of multiple components including
CPU, Graphics, LAN, Disk I/O and Compiler Technology. As workstations
cross the 40-50 MIPs boundary, disk I/O becomes an important factor to
achieving balanced workstation performance (Slide 1).
At HP's Fort Collins Systems Division, disk I/O behavior for
workstation applications is measured and analyzed. These applications
areas include 2D Drafting, 3D Mechanical Design and CASE, examined in
both stand-alone and client-server configurations. Seeing how disks
behave while running various applications helps us to determine what
disk characteristics will benefit customers' application performance
most.
The Fort Collins data suggests that most disk benchmarks measure
parameters that account for less than 15% of real disk activity. The
industry has been introducing disks with very good performance on large
file transfers where as we find that accessing random, small files of
data accounts for 85-95% of all disk I/O activity (Slide 3). At 40
Random I/O's per second, HP's 330S/660S offer the industrys best Random
Performance.
The data also suggests that disk "Read" and "Write" behavior occur in
equal frequency when running these applications (Slide 2). HP's
330S/660S again should run faster for these applications since they
offer equivalent "Read" and "Write" performance. Most OEM SCSI disk
manufacturers (Maxtor, Connor, Imprimis) used by SUN and DEC have far
better "Read" than "Write" performance. In many cases, Sun won't even
publish the "Write" performance for their OEM disks.
Sun's IPI-2 solution delivers good performance on the sequential
transfer of large files but does little to improve the performance of
random data access. HP's 330S/660S disks offer slightly better random
performance to Sun's IPI disks at 50% of the price/Megabyte.
Therefore, by buying two SCSI disks instead of one expensive IPI disk
the customer can improve performance and save money.
In conclusion, what appears to matter most for the applications we
tested was random performance. As mentioned on WKSG, Sun's IPI
disk/controller products have very little impact on random performance.
To improve random performance (Slide 4) use multiple disks (e.g. do VM
paging on a separate disk) which will also improve application
performance (Slide 4).