Finer reference trace

.github/configs/wordlist.txt

@@ -748,4 +748,6 @@ Lazar
 Cvetkovic
 cvetkovic
 ethz
-lazar
+lazar
+xvzf
+untar

.github/workflows/integration_tests.yaml

@@ -43,7 +43,7 @@ jobs:
     - name: Drawing samples
       run: |
         tar -xzvf $tpath/inputs/preprocessed.tar.gz -C $tpath/inputs/
-        python -m sampler sample --source_trace $tpath/inputs/preprocessed --output $tpath/sampled --min-size 10 --step-size=10 --max-size=50
+        python -m sampler sample --source_trace $tpath/inputs/preprocessed --original_trace $tpath/inputs/preprocessed --output $tpath/sampled --min-size 10 --step-size=10 --max-size=50
 
 #    - name: Plotting results
 #      run: |

.gitignore

@@ -5,10 +5,7 @@ analysis
 tmp
 data/out
 data/azure
-data/traces/*
 !data/traces/example/
-data/traces/reference/*/*.csv
-!data/traces/reference/
 pkg/generator/*.png
 pkg/generator/*.txt
 pkg/driver/*.csv

data/traces/reference/.gitignore

@@ -0,0 +1 @@
+*.csv

docs/loader.md

@@ -124,6 +124,10 @@ For to configure the workload for load generator, please refer to `docs/configur
 There are a couple of constants that should not be exposed to the users. They can be examined and changed
 in `pkg/common/constants.go`.
 
+Sample sizes appropriate for performance evaluation vary depending on the platform. 
+As a starting point for fine-tuning, we suggest at most 5 functions per core with SMT disabled. 
+For example, 80 functions for a 16-core node. With larger sample sizes, trace replaying may lead to failures in function invocations.
+
 ## Build the image for a synthetic function
 
 The reason for existence of Firecracker and container version is because of different ports for gRPC server. Firecracker

docs/sampler.md

@@ -12,7 +12,7 @@ git lfs install
 cd sampler
 git lfs fetch
 git lfs checkout
-pip install -r requirements.txt
+pip install -r ../requirements.txt
 ```
 
 ## Pre-processing the original trace (mandatory)
@@ -91,9 +91,9 @@ monotonic load increase (in terms of resource usage) when sweeping the sample si
 ```console
 python3 -m sampler sample -h
 
-usage: sample [-h] -t path -o path [-min integer] [-st integer] [-max integer] [-tr integer]
+usage: sample [-h] -t path -orig path -o path [-min integer] [-st integer] [-max integer] [-tr integer]
 
-optional arguments:
+options:
   -h, --help            show this help message and exit
   -t path, --source_trace path
                         Path to trace to draw samples from
@@ -113,22 +113,31 @@ optional arguments:
 
 ## Reference traces
 
-The reference traces are stored in `data/traces/reference` folder of this repository, as `preprocessed.tar.gz` and
-`sampled.tar.gz` files stored in Git LFS.
+The reference traces are stored in `data/traces/reference` folder of this repository, as `preprocessed_150.tar.gz` and
+`sampled_150.tar.gz` files stored in Git LFS.
+
+`preprocessed_150.tar.gz` contains the preprocessed traces for the original Azure trace for day 1, 09:00:00-11:30:00 (150
+minutes total). 150 minutes trace captures approximately half of all functions from original Azure trace, but makes it
+more suitable to run in shorter experiments (10 minutes - 2 hours).
+
+`sampled_150.tar.gz` contains the sampled traces for preprocessed trace from `preprocessed_150.tar.gz`. Sample sizes are
+10-200 functions with step 10, 200-3k with step 50, and 3k-24k with step 1k.
 
-`preprocessed.tar.gz` contains the preprocessed traces for the original Azure trace for day 1, 09:00:00-11:30:00 (150
-minutes total).
+You can untar the tarballs with the following commands:
 
-`sampled.tar.gz` contains the sampled traces for preprocessed trace from `preprocessed.tar.gz`. Sample sizes are 50-3k
-functions with step 50 and 3k-24k with step 1k.
+```console
+tar -xvzf sampled_150.tar.gz
+tar -xvzf preprocessed_150.tar.gz
+```
 
 The reference traces were obtained by running the following commands:
 
 ```console
-python3 -m preprocess  -t data/azure/ -o data/reference/preprocessed_150 -s 00:09:00 -dur 150
+python3 -m sampler preprocess  -t data/azure/ -o data/traces/reference/preprocessed_150 -s 00:09:00 -dur 150
 
-python3 -m sample -t data/reference/preprocessed_150 -o data/reference/sampled_150 -min 3000 -st 1000 -max 24000 -tr 16
-python3 -m sample -t data/reference/sampled_150/samples/3000 -o data/reference/sampled_150 -min 50 -st 50 -max 3000 -tr 16
+python3 -m sampler sample -t data/traces/reference/preprocessed_150 -orig data/traces/reference/preprocessed_150 -o data/traces/reference/sampled_150 -min 3000 -st 1000 -max 24000 -tr 16
+python3 -m sampler sample -t data/traces/reference/sampled_150/samples/3000 -orig data/traces/reference/preprocessed_150 -o data/traces/reference/sampled_150 -min 200 -st 50 -max 3000 -tr 16
+python3 -m sampler sample -t data/traces/reference/sampled_150/samples/200 -orig data/traces/reference/preprocessed_150 -o data/traces/reference/sampled_150 -min 10 -st 10 -max 200 -tr 16
 ```
 
 ## Tools

requirements.txt

@@ -1,7 +1,7 @@
 matplotlib==3.7.2
-numpy==1.26.1
+numpy==1.24.4
 pandas==1.3.5
-scipy==1.11.2
+scipy==1.10.1
 pytest==7.4.0
 cloudpickle==2.2.1
 seaborn==0.13.0

sampler/__main__.py

@@ -108,6 +108,7 @@ def main():
     sample_parser.add_argument(
         '-orig',
         '--original_trace',
+        required=True,
         metavar='path',
         default=None,
         help='Path to the Azure (or other original) trace files, required to maximize the derived sample\'s representativity (WD from the original trace)'