lcs.64k.fio

[global]
rw=readwrite                        # type of load wanted
ioengine=libaio                     # libaio = best for linux, mmap if not supported (Mac)
iodepth=8                           # baseline of a medium used server, increase to 16-32 for more heavily to max used
invalidate=1                        # Invalidate the buffer/page cache parts for this file prior to starting I/O
direct=1                            # bypass os buffers
randrepeat=1                        # run one pass of the defined time
fallocate=none                      # do not pre-allocate space when files are laid down
thread=1                            # only run one thread for each job
fadvise_hint=0                      # don't alert the OS kernel of type of load
time_based=1                        # run on timer vs. rounds
directory=./data                    # directory to store files
file_service_type=sequential        # finish one file write then move to the next
filename_format=cassandra.$filenum  # create named files to be reused in all jobs below
nrfiles=10                          # number of files to read/write
filesize=160m                       # LCS target is 160mb
blocksize=64k                       # default sstable compaction chunk size


[setup]
name=lcs_setup      # name in end report
readwrite=write     # write only load type
end_fsync=1         # on finish force fsync of dirty bytes to disk
runtime=1           # just create the base files and move on
stonewall           # block on this job until all files are created


[sstable_writer]
new_group                       # create a new grouping in the report to separate setup and real load tests
name=lcs_64k_write              # name in end report
readwrite=write                 # write only job
runtime=60s                     # run for 60s
openfiles=1                     # simulate single threaded compaction
write_bw_log=lcs.64k.write      # output throughput results to graph
write_lat_log=lcs.64k.write     # output latency results to graph
write_iops_log=lcs.64k.write    # output iops results to graph


[sstable_reader]
name=lcs_64k_read               # name in end report
rw=randread                     # perform random reads
runtime=60s                     # run for 60s
openfiles=5                     # simulate random reads across multiple sstables
file_service_type=random        # randomly read across all files
write_bw_log=lcs.64k.read       # output throughput results to graph
write_lat_log=lcs.64k.read      # output latency results to graph
write_iops_log=lcs.64k.read     # output iops results to graph