AIS Loader

AIS Loader (aisloader) is a tool to measure storage performance. It is a load generator that we constantly use to benchmark and stress-test AIStore or any S3-compatible backend.

In fact, aisloader can list, write, and read S3() buckets directly, which makes it quite useful, convenient, and easy to use benchmark to compare storage performance **with aistore in front of S3 and without.

(**) aisloader can be further easily extended to work directly with any Cloud storage provider including, but not limited to, aistore-supported GCP and Azure.

In addition, aisloader generates synthetic workloads that mimic training and inference workloads - the capability that allows to run benchmarks in isolation (which is often preferable) avoiding compute-side bottlenecks (if any) and associated complexity.

There’s a large set of command-line switches that allow to realize almost any conceivable workload, with basic permutations always including:

  • number of workers
  • read and write sizes
  • read and write ratios

Detailed protocol-level tracing statistics are also available - see HTTP tracing section below for brief introduction.

Table of Contents


To get started, go to root directory and run:

$ make aisloader

For usage, run: aisloader or aisloader usage or aisloader --help.

Command-line Options

For the most recently updated command-line options and examples, please run aisloader or aisloader usage.

Options via AIS Loader flags

Command-line option Type Description Default
-batchsize int Batch size to list and delete 100
-bprops json JSON string formatted as per the SetBucketProps API and containing bucket properties to apply ""
-bucket string Bucket name. Bucket will be created if doesn’t exist. If empty, aisloader generates a new random bucket name ""
-cached bool list in-cluster objects - only those objects from a remote bucket that are present (“cached”) false
-cksum-type string Checksum type to use for PUT object requests xxhash
-cleanup bool when true, remove bucket upon benchmark termination n/a (required)
-dry-run bool show the entire set of parameters that aisloader will use when actually running false
-duration string, int Benchmark duration (0 - run forever or until Ctrl-C, default 1m). Note that if both duration and totalputsize are zeros, aisloader will have nothing to do 1m
-epochs int Number of “epochs” to run whereby each epoch entails full pass through the entire listed bucket 1
-etl string Built-in ETL, one-of: tar2tf, md5, or echo. Each object that aisloader GETs undergoes the selected transformation. See also: -etl-spec option. ""
-etl-spec string Custom ETL specification (pathname). Must be compatible with Kubernetes Pod specification. Each object that aisloader GETs will undergo this user-defined transformation. See also: -etl option. ""
-getconfig bool when true, generate control plane load by reading AIS proxy configuration (that is, instead of reading/writing data exercise control path) false
-getloaderid bool when true, print stored/computed unique loaderID aka aisloader identifier and exit false
-ip string AIS proxy/gateway IP address or hostname localhost
-json bool when true, print the output in JSON false
-loaderid string ID to identify a loader among multiple concurrent instances 0
-loaderidhashlen int Size (in bits) of the generated aisloader identifier. Cannot be used together with loadernum 0
-loadernum int total number of aisloaders running concurrently and generating combined load. If defined, must be greater than the loaderid and cannot be used together with loaderidhashlen 0
-maxputs int Maximum number of objects to PUT 0
-maxsize int Maximal object size, may contain multiplicative suffix 1GiB
-minsize int Minimal object size, may contain multiplicative suffix 1MiB
-numworkers int Number of goroutine workers operating on AIS in parallel 10
-pctput int Percentage of PUTs in the aisloader-generated workload 0
-latest bool When true, check in-cluster metadata and possibly GET the latest object version from the associated remote bucket false
-port int Port number for proxy server 8080
-provider string ais - for AIS, cloud - for Cloud bucket; other supported values include “gcp” and “aws”, for Amazon and Google clouds, respectively ais
-putshards int Spread generated objects over this many subdirectories (max 100k) 0
-quiet bool When starting to run, do not print command line arguments, default settings, and usage examples false
-randomname bool when true, generate object names of 32 random characters. This option is ignored when loadernum is defined true
-readertype string Type of reader: sg(default). Available: sg, file, rand, tar sg
-readlen string, int Read range length, can contain multiplicative suffix ""
-readoff string, int Read range offset (can contain multiplicative suffix K, MB, GiB, etc.) ""
-s3endpoint string S3 endpoint to read/write S3 bucket directly (with no aistore) ""
-s3profile string Other then default S3 config profile referencing alternative credentials ""
-seed int Random seed to achieve deterministic reproducible results (0 - use current time in nanoseconds) 0
-skiplist bool Whether to skip listing objects in a bucket before running PUT workload false
-filelist string Local or locally accessible text file file containing object names (for subsequent reading) ""
-stats-output string filename to log statistics (empty string translates as standard output (default) ""
-statsdip string StatsD IP address or hostname localhost
-statsdport int StatsD UDP port 8125
-statsdprobe bool Test-probe StatsD server prior to benchmarks true
-statsinterval int Interval in seconds to print performance counters; 0 - disabled 10
-subdir string Virtual destination directory for all aisloader-generated objects ""
-test-probe bool Test StatsD server prior to running benchmarks false
-timeout string Client HTTP timeout; 0 = infinity) 10m
-tmpdir string Local directory to store temporary files /tmp/ais
-tokenfile string Authentication token (FQN) ""
-totalputsize string, int Stop PUT workload once cumulative PUT size reaches or exceeds this value, can contain multiplicative suffix, 0 = no limit 0
-trace-http bool Trace HTTP latencies (see HTTP tracing) false
-uniquegets bool when true, GET objects randomly and equally. Meaning, make sure not to GET some objects more frequently than the others true
-usage bool Show command-line options, usage, and examples false
-verifyhash bool checksum-validate GET: recompute object checksums and validate it against the one received with the GET metadata true

Often used options explanation


The loads can run for a given period of time (option -duration <duration>) or until the specified amount of data is generated (option -totalputsize=<total size in KBs>).

If both options are provided the test finishes on the whatever-comes-first basis.

Example 100% write into the bucket “abc” for 2 hours:

$ aisloader -bucket=abc -provider=ais -duration 2h -totalputsize=4000000 -pctput=100

The above will run for two hours or until it writes around 4GB data into the bucket, whatever comes first.

Write vs Read

You can choose a percentage of writing (versus reading) by setting the option -pctput=<put percentage>.

Example with a mixed PUT=30% and GET=70% load:

$ aisloader -bucket=ais://abc -duration 5m -pctput=30 -cleanup=true

Example 100% PUT:

$ aisloader -bucket=abc -duration 5m -pctput=100 -cleanup=true

The duration in both examples above is set to 5 minutes.

To test 100% read (-pctput=0), make sure to fill the bucket beforehand.

Read range

The loader can read the entire object (default) or a range of object bytes.

To set the offset and length to read, use option -readoff=<read offset (in bytes)> and readlen=<length to read (in bytes)>.

For convenience, both options support size suffixes: k - for KiB, m - for MiB, and g - for GiB.

Example that reads a 32MiB segment at 1KB offset from each object stored in the bucket “abc”:

$ aisloader -bucket=ais://abc -duration 5m -cleanup=false -readoff=1024 -readlen=32m

The test (above) will run for 5 minutes and will not “cleanup” after itself (next section).


NOTE: -cleanup is a mandatory option defining whether to destroy bucket upon completion of the benchmark.

The option must be specified in the command line.


$ aisloader -bucket=ais://abc -pctput=100 -totalputsize=16348 -cleanup=false
$ aisloader -bucket=ais://abc -duration 1h -pctput=0 -cleanup=true

The first line in this example above fills the bucket “abc” with 16MiB of random data. The second - uses existing data to test read performance for 1 hour, and then removes all data.

If you just need to clean up old data prior to running a test, run the loader with 0 (zero) total put size and zero duration:

$ aisloader -bucket=<bucket to cleanup> -duration 0s -totalputsize=0

Object size

For the PUT workload the loader generates randomly-filled objects. But what about object sizing?

By default, object sizes are randomly selected as well in the range between 1MiB and 1GiB. To set preferred (or fixed) object size(s), use the options -minsize=<minimal object size in KiB> and -maxsize=<maximum object size in KiB>

Setting bucket properties

Before starting a test, it is possible to set mirror or EC properties on a bucket (for background, please see storage services).

For background on local mirroring and erasure coding (EC), please see storage services.

To achieve that, use the option -bprops. For example:

$ aisloader -bucket=ais://abc -pctput=0 -cleanup=false -duration 10s -bprops='{"mirror": {"copies": 2, "enabled": false}, "ec": {"enabled": false, "data_slices": 2, "parity_slices": 2}}'

The above example shows the values that are globally default. You can omit the defaults and specify only those values that you’d want to change. For instance, to enable erasure coding on the bucket “abc”:

$ aisloader -bucket=ais://abc -duration 1h -bprops='{"ec": {"enabled": true}}' -cleanup=false

This example sets the number of data and parity slices to 2 which, in turn, requires the cluster to have at least 5 target nodes: 2 for data slices, 2 for parity slices and one for the original object.

Once erasure coding is enabled, its properties data_slices and parity_slices cannot be changed on the fly.

Note that (n data_slices, m parity_slices) erasure coding requires at least (n + m + 1) target nodes in a cluster.

Even though erasure coding and/or mirroring can be enabled/disabled and otherwise reconfigured at any point in time, specifically for the purposes of running benchmarks it is generally recommended to do it once prior to writing any data to the bucket in question.

The following sequence populates a bucket configured for both local mirroring and erasure coding, and then reads from it for 1h:

# Fill bucket
$ aisloader -bucket=ais://abc -cleanup=false -pctput=100 -duration 100m -bprops='{"mirror": {"enabled": true}, "ec": {"enabled": true}}'

# Read
$ aisloader -bucket=abc -cleanup=false -pctput=0 -duration 1h

Bytes Multiplicative Suffix

Parameters in aisLoader that represent the number of bytes can be specified with a multiplicative suffix. For example: 8M would specify 8 MiB. The following multiplicative suffixes are supported: ‘t’ or ‘T’ - TiB ‘g’ or ‘G’ - GiB, ‘m’ or ‘M’ - MiB, ‘k’ or ‘K’ - KiB. Note that this is entirely optional, and therefore an input such as 300 will be interpreted as 300 Bytes.

Environment variables

Environment Variable Type Description
AIS_ENDPOINT string Cluster’s endpoint: http or https address of any aistore gateway in this cluster. Overrides ip and port flags.

To state the same slightly differently, cluster endpoint can be defined in two ways:

  • as (plain) http://ip:port address, whereby ‘--ip’ and ‘--port’ are command-line options.
  • via AIS_ENDPOINT environment universally supported across all AIS clients, e.g.:
$ export AIS_ENDPOINT=

In addition, environment can be used to specify client-side TLS (aka, HTTPS) configuration:

var name description
AIS_CRT X509 certificate
AIS_CRT_KEY X509 certificate’s private key
AIS_CLIENT_CA Certificate authority that authorized (signed) the certificate
AIS_SKIP_VERIFY_CRT when true, skip X509 cert verification (usually enabled to circumvent limitations of self-signed certs)


For the most recently updated command-line options and examples, please run aisloader or aisloader usage.

1. Create a 10-seconds load of 50% PUT and 50% GET requests:

$ aisloader -bucket=my_ais_bucket -duration=10s -pctput=50 -provider=ais
Found 0 existing objects
Run configuration:
    "proxy": "",
    "provider": "ais",
    "bucket": "my_ais_bucket",
    "duration": "10s",
    "put upper bound": 0,
    "put %": 50,
    "minimal object size in Bytes": 1024,
    "maximum object size in Bytes": 1048576,
    "worker count": 1,
    "stats interval": "10s",
    "backed by": "sg",
    "cleanup": true

Actual run duration: 10.313689487s

Time      OP    Count                 	Total Bytes           	Latency(min, avg, max)              	Throughput            	Error
01:52:52  Put   26                    	11.19GB               	296.39ms   5.70s      14.91s        	639.73MB              	0
01:52:52  Get   16                    	3.86GB                	58.89ms    220.20ms   616.72ms      	220.56MB              	0
01:52:52  CFG   0                     	0B                    	0.00ms     0.00ms     0.00ms        	0B                    	0
01:52:52 Clean up ...
01:52:54 Clean up done

2. Time-based 100% PUT into ais bucket. Upon exit the bucket is destroyed:

$ aisloader -bucket=nvais -duration 10s -cleanup=true -numworkers=3 -minsize=1K -maxsize=1K -pctput=100 -provider=ais

3. Timed (for 1h) 100% GET from a Cloud bucket, no cleanup:

$ aisloader -bucket=aws://nvaws -duration 1h -numworkers=30 -pctput=0 -cleanup=false

4. Mixed 30%/70% PUT and GET of variable-size objects to/from a Cloud bucket. PUT will generate random object names and is limited by the 10GB total size. Cleanup enabled - upon completion all generated objects and the bucket itself will be deleted:

$ aisloader -bucket=s3://nvaws -duration 0s -cleanup=true -numworkers=3 -minsize=1024 -maxsize=1MB -pctput=30 -totalputsize=10G

5. PUT 1GB total into an ais bucket with cleanup disabled, object size = 1MB, duration unlimited:

$ aisloader -bucket=nvais -cleanup=false -totalputsize=1G -duration=0 -minsize=1MB -maxsize=1MB -numworkers=8 -pctput=100 -provider=ais

6. 100% GET from an ais bucket:

$ aisloader -bucket=nvais -duration 5s -numworkers=3 -pctput=0 -provider=ais -cleanup=false

7. PUT 2000 objects named as aisloader/hex({0..2000}{loaderid}):

$ aisloader -bucket=nvais -duration 10s -numworkers=3 -loaderid=11 -loadernum=20 -maxputs=2000 -objNamePrefix="aisloader" -cleanup=false

8. Use random object names and loaderID to report statistics:

$ aisloader -loaderid=10

9. PUT objects with random name generation being based on the specified loaderID and the total number of concurrent aisloaders:

$ aisloader -loaderid=10 -loadernum=20

10. Same as above except that loaderID is computed by the aisloader as hash(loaderstring) & 0xff:

$ aisloader -loaderid=loaderstring -loaderidhashlen=8

11. Print loaderID and exit (all 3 examples below) with the resulting loaderID shown on the right:

$ aisloader -getloaderid (0x0)
$ aisloader -loaderid=10 -getloaderid (0xa)
$ aisloader -loaderid=loaderstring -loaderidhashlen=8 -getloaderid (0xdb)

12. Destroy existing ais bucket. If the bucket is Cloud-based, delete all objects:

$ aisloader -bucket=nvais -duration 0s -totalputsize=0 -cleanup=true

13. Generate load on a cluster listening on custom IP address and port:

$ aisloader -ip="" -port=8080

14. Generate load on a cluster listening on custom IP address and port from environment variable:

$ AIS_ENDPOINT="" aisloader

15. Use HTTPS when connecting to a cluster:

$ aisloader -ip="https://localhost" -port=8080

16. PUT TAR files with random files inside into a cluster:

$ aisloader -bucket=my_ais_bucket -duration=10s -pctput=100 -provider=ais -readertype=tar

17. Generate load on tar2tf ETL. New ETL is started and then stopped at the end. TAR files are PUT to the cluster. Only available when cluster is deployed on Kubernetes.

$ aisloader -bucket=my_ais_bucket -duration=10s -pctput=100 -provider=ais -readertype=tar -etl=tar2tf -cleanup=false

18. Timed 100% GET directly from S3 bucket (notice ‘-s3endpoint’ command line): console $ aisloader -bucket=s3://xyz -cleanup=false -numworkers=8 -pctput=0 -duration=10m -s3endpoint=

19. PUT approx. 8000 files into s3 bucket directly, skip printing usage and defaults. Similar to the previous example, aisloader goes directly to a given S3 endpoint (‘-s3endpoint’), and aistore is not being used: console $ aisloader -bucket=s3://xyz -cleanup=false -minsize=16B -maxsize=16B -numworkers=8 -pctput=100 -totalputsize=128k -s3endpoint= -quiet

20. Generate a list of object names (once), and then run aisloader without executing list-objects:

$ ais ls ais://nnn --props name -H > /tmp/a.txt
$ aisloader -bucket=ais://nnn -duration 1h -numworkers=30 -pctput=0 -filelist /tmp/a.txt -cleanup=false

Collecting stats

Collecting is easy - aisloader supports at-runtime monitoring via with Graphite using StatsD. When starting up, aisloader will try to connect to provided StatsD server (see: statsdip and statsdport options). Once the connection is established the statistics from aisloader are send in the following format:

  • metric_type - can be: gauge, timer, counter
  • hostname - is the hostname of the machine on which the loader is ran
  • loaderid - see: -loaderid option
  • metric - can be: latency.*, get.*, put.* (see: aisloader metrics)


Grafana helps visualize the collected statistics. It is convenient to use and provides numerous tools to measure and calculate different metrics.

We provide simple script which allows you to set up the Graphite and Grafana servers which run inside separate dockers. To add new dashboards and panels, please follow: grafana tutorial.

When selecting a series in panel view, it should be in the format: stats.aisloader.<loader>.*. Remember that metrics will not be visible (and you will not be able to select them) until you start the loader.

HTTP tracing

Following is a brief illustrated sequence to enable detailed tracing, capture statistics, and toggle tracing on/off at runtime.


The amount of generated (and extremely detailed) metrics can put a strain on your StatsD server. That’s exactly the reason for runtime switch to toggle HTTP tracing on/off. The example below shows how to do it (in particular, see kill -HUP).

1. Run aisloader for 90s (32 workes, 100% write, sizes between 1KB and 1MB) with detailed tracing enabled:

$ aisloader -bucket=ais://abc -duration 90s -numworkers=32 -minsize=1K -maxsize=1M -pctput=50 --cleanup=false --trace-http=true

2. Have netcat listening on the default StatsD port 8125:

$ nc 8125 -l -u -k

# The result will look as follows - notice "*latency*" metrics (in milliseconds):


3. Finally, toggle detailed tracing on and off by sending aisoader SIGHUP:

$ pgrep -a aisloader
3800 aisloader -bucket=ais://abc -duration 90s -numworkers=32 -minsize=1K -maxsize=1M -pctput=100 --cleanup=false --trace-http=true

# kill -1 3800
# or, same: kill -HUP 3800

The result:

Time      OP    Count                   Size (Total)            Latency (min, avg, max)                 Throughput (Avg)        Errors (Total)
10:11:27  PUT   20,136 (20,136 8 0)     19.7MiB (19.7MiB)       755.308µs  3.929ms    42.493ms         1.97MiB/s (1.97MiB/s)   -
Detailed latency info is disabled

# As stated, `SIGHUP` is a binary toggle - next time used it'll enable detailed trace with `aisloader printing:

Detailed latency info is enabled

Note that other than --trace-http, all command-line options in this section are used for purely illustrative purposes.

AISLoader Composer

For benchmarking production-level clusters, a single AISLoader instance may not be able to fully saturate the load the cluster can handle. In this case, multiple aisloader instances can be coordinated via the AISLoader Composer. See the README for instructions on setting up.


For documented aisloader metrics, please refer to:

The same readme (above) also describes:

For StatsD compliant backends, see:

Finally, for another supported - and alternative to StatsD - monitoring via Prometheus integration, see: