S3Bench

S3Bench is a light-weight S3 service benchmark tool, for measuring 
latencies and throughput for object up- and downloads. This software 
uses the AWS S3 Java SDK as foundation and provides basic CLI and 
properties file-based customization for benchmark workloads.


Getting started

>./s3bench.sh s3.domain.com mybucket accessKeyId secret*** [s3bench.properties]


Customization options (s3bench.properties)

The given values match the fallback default properties, if the optional 
properties file command line argument was not specified.

# Determines whether requests use secure or plain-text connections.
# true: requests are using HTTPS (verifiable certificates only)
# false: plain-text HTTP requests
use_ssl=false

# Specifies how buckets are encoded in the client request URLs.
# true: via URL path (i.e. http://s3.domain.com/mybucket)
# false: via sub-domain (i.e. http://mybucket.s3.domain.com)
path_style=true

# MD5 checksum verification by the client prolong the overall request latency.
# true: verification is enabled
# false: verification is disabled
client_md5_verification=false

# If the bucket retrieved as a command line option is not existent this flag 
# determines, whether this tool may create it.
# true: s3bench may create the target bucket
# false: s3bench will fail if the target bucket does not exist
create_bucket=true

# The number of objects created during a single stream benchmark.
sample_count=10

# Object payload is transmitted unsigned, removing the need of server side 
# checksum generation (fast).
benchmark_unsigned=true

# Object payload is signed (and verified at the server) via AWS S3 signature V4 
# chunked uploads.
benchmark_chunked=true

# Object payload is SHA256 pre-signed and verified by the server.
benchmark_presigned=true

# Any objects in the target bucket are removed after the run.
# true: remove _any_ objects in the target bucket
# false: leave the bucket as is after finishing the benchmark
clean_after_run=true

# Uses large objects to obtain throughput key figures.
# true: run throughput workloads
# false: skip throughput workloads 
benchmark_tp=true

# Object size in bytes during throughput benchmarks in bytes (limit: < 2^31).
tp_object_size=1073741824

# Uses small objects to obtain latency key figures.
# true: run latency workloads
# false: skip latency workloads
benchmark_ops=true

# Object size in bytes during latency benchmarks in bytes (limit: < 2^31).
ops_object_size=4096

# Run single-stream workloads for sequential performance benchmarks.
# true: run sequential workloads
# false: skip sequential workloads
benchmark_single_stream=true

# Run multi-stream workloads for parallel performance benchmarks.
# true: run parallel workloads
# false: skip parallel workloads
benchmark_multi_stream=true

# The runtime of a single stream benchmark in seconds.
multi_stream_runtime_s=600

# Number of clients during a multi-stream throughput benchmark.
multi_stream_tp_clients=2

# Number of threads per client during a multi-stream throughput benchmark.
multi_stream_tp_threads=10

# Number of clients during multi-stream latency benchmarks.
multi_stream_ops_clients=10

# Number of threads per client during multi-stream latency benchmarks.
multi_stream_ops_threads=10

# Run download object benchmarks after object upload workloads.
# true: run download object workloads
# false: skip download object workloads
benchmark_get=true