I/O Workload

The I/O workload is designed to simulate users performing both production and development runs of I/O-intensive applications. Each workload was designed to run for between 8 and 12 minutes on a single processor Sparc Station 20.

The workloads can be grouped into three categories, according to the dependencies across the I/O that they perform. Some of the users run applications which produce large output files, in which case we simulate the user examiniming the output. Other users run applications that continually use the same input files. Finally, other users run a series of applications, with the output of one being the input of another.

The simulated users run the following programs, performing the indicated amount of I/O.

Seis: Part of the ParkBench Compact Application Suite

seis creates 1.5 MB file
sbfdump reads that 1.5 MB file
seis reads that 1.5 MB file; creates 128K file
sbfdump reads that 128 K file

seis creates 1.5 MByte
seis reads 1.5 Mbyte; creates 128K file
sbfdump reads 1.5 MByte; creates 100K
sbfdump reads 128 K MByte; creates 18K

seis creates 8 MByte file
sbfdump reads 8 MByte; creates 600K
seis reads 8 Mbyte; creates 128K
sbfdump reads 128 K file

seis creates 8 MByte
seis reads 8 Mbyte; creates 128K file
sbfdump reads 8 MByte; creates 600K
sbfdump reads 128 K MByte

Arps:
One run of Arps creates a 15 MB file
User parses this output

mpeg:
Unzips 5 files of 3.2 MB each; creates 5 3.8 MB files
Untar of these 5 files creates 150 120K files
mpeg_encode of 150 files
mpeg_play on result

schedsim1:
User alternates between performing small runs of simulator and
parsing output with awk

sim1:
20 small runs, each reads the same large input files

out-of-core sort: Alternates between the following:
Creates large file of keys
Sorts those keys and writes out result

sim2:
3 large runs, each reads the same input files

schedsim2:
User alternates between performing large runs of simulator and
parsing output with awk