Adding instructions for mini jobs to README files (#160)

Summary:

This diff adss instructions for running Spark, Django, and Mediawiki mini job

Differential Revision: D78453571

Author: marziehlenjaniMeta

packages/django_workload/README.md

@@ -78,6 +78,53 @@ If running on ARM platform, please use the job `django_workload_arm`:
 ./benchpress_cli.py run django_workload_arm -r standalone
 ```
 
+### Run DjangoBench Mini
+
+DjangoBench Mini is a shrunken version of DjangoBench that
+aims to reduce execution time to less than 30 seconds
+and can potentially be used for emulations.
+It reuses the dataset generated by the full DjangoBench and has a shorter runtime because
+ it runs only one iteration with the number of repetitions set to 100.
+
+To run DjangoBench Mini, please follow these steps:
+
+1. Make sure that you get the latest version of DCPerf and check out the latest commit in the `v2-beta` branch.
+If you've installed DjangoBench with an older version of DCPerf,
+we recommend you clean and re-install DjangoBench.
+
+2. Run the regular version of DjangoBench on a real machine for the first time to build the dataset
+
+- If you are using an ARM machine, run the `django_workload_arm` job:
+
+```
+./benchpress run django_workload_arm -r standalone
+```
+
+- On x86 machines, run the `django_workload_default` job:
+
+```
+./benchpress run django_workload_default -r standalone
+```
+
+This will generates the necessary dataset and stores it in /data/cassandra/data.
+Create a backup of this folder to avoid data loss.
+
+3. Copy over `/data/cassandra/data` to the servers or the emulator that you would like to benchmark, then run DjangoBench mini:
+
+- For ARM:
+
+```
+./benchpress run django_workload_arm_mini -r standalone
+```
+
+- For x86:
+
+```
+./benchpress run django_workload_mini -r standalone
+```
+
+These mini jobs will reuse the dataset generated in the previous step.
+
 ## Reporting
 
 Once the benchmark finishes on the django benchmarking machine, benchpress will

packages/mediawiki/README.md

@@ -133,6 +133,41 @@ you can run the following:
 ./benchpress_cli.py run oss_performance_mediawiki_mlp -i '{"scale_out": 3}'
 ```
 
+### Running Mediawiki mini
+
+Mediawiki mini is a shrunken version of Mediawiki
+ that aims to reduce execution time to less than two minutes.
+ Mediawiki mini includes a few new options that
+  allow user to customize the benchmarking process:
+
+- Reuse the temporary folder between runs.
+- Adjust the load generator seed.
+- Configure load generation durations for warmup.
+- Set the interval for JIT retranslation.
+- Specify a deterministic number of warmup iterations.
+
+To run the mini version of
+Mediawiki, please follow these steps:
+
+1. Make sure that you get the latest version of DCPerf and check out
+the latest commit in the `v2-beta` branch.
+If you've installed Mediawiki with an older version of DCPerf,
+we recommend that you clean and re-install Mediawiki.
+
+2. Run the `oss_performance_mediawiki_mini` job with the following command:
+
+```bash
+./benchpress run oss_performance_mediawiki_mini
+```
+
+3. To further reduce execution time by reusing the temporary directory, please run the
+`oss_performance_mediawiki_mini` job once, locate the path to the temp folder in the log output
+(named like `/tmp/hhvm-nginxXXXXX`), then specify `temp_dir` parameter in the subsequent run:
+
+```bash
+./benchpress run oss_performance_mediawiki_mini -i '{"temp_dir":"/tmp/hhvm-nginxtIdDIr"}'
+```
+
 ### Reporting
 
 After the benchmark finishes, benchpress will report the benchmark results in the

packages/spark_standalone/README.md

@@ -372,6 +372,53 @@ faster with the already built database. Therefore, it's normal that the first ru
 takes much longer time than what the benchmark eventually reports. The database will
 be stored in `/flash23/warehouse`.
 
+### Running SparkBench Mini
+
+SparkBench Mini is a shrunken version of SparkBench that aims to reduce execution time to
+less than 30 seconds and can potentially be used for emulations. To run the mini version of
+SparkBench, please follow the following steps:
+
+1. Download one of the down-sampled version of the dataset:
+    - 5GB variant `bpc_t93586_s2_synthetic_5GB`
+    - 1GB variant `bpc_t93586_s2_synthetic_1GB`
+
+You can download these datasets from the following Github repository:
+
+```
+https://github.com/facebookresearch/DCPerf-datasets
+```
+
+2. Make sure that you get the latest version of DCPerf and check out the latest commit in the `v2-beta` branch.
+If you've installed SparkBench with an older version of DCPerf,
+we recommend you clean and re-install SparkBench.
+
+3. If you have run SparkBench with a different dataset,
+please remove the data from previous runs so that SparkBench can rebuild database with the new dataset:
+
+```bash
+rm -rf /flash23/warehouse
+rm -rf <benchpressPath>/benchmarks/spark_standalone/spark-2.4.5-bin-hadoop2.7/metastore_db
+```
+
+4. Create the `/flash23` folder. Copy `bpc_t93586_s2_synthetic_5GB`
+or `bpc_t93586_s2_synthetic_1GB` in the `/flash23` folder.
+Note that SparkBench mini does not require the high I/O throughput
+ as the regular version, so it's OK to put `/flash23` folder on your system drive.
+
+5. Run `spark_standalone_remote_mini` job on a real machine.
+This will create data in `/flash23/warehouse`
+and `<benchpressPath>/benchmarks/spark_standalone/spark-2.4.5-bin-hadoop2.7/metastore_db`.
+Create a backup of these two folders. By default, this job uses the 5GB dataset.
+If you want to use the 1GB dataset, run the job with specifying the
+`dataset_name` parameter like this:
+
+```bash
+./benchpress run spark_standalone_remote_mini -i '{"dataset_name":"bpc_t93586_s2_synthetic_1GB"}'
+```
+
+6. Run Spark mini on a real machine/emulator for the second time to reuse data,
+with the same commands and options.
+
 ## Reusing database on another machine
 
 Building database takes a considerable amount of time, so it's advisable to consider