This repository provides the instructions and files required to run the simulations and extract the results.We use Omnet++ simulator and INET framework to run the simulations. Omnet++ manual and its examples are good references for introduction to the simulator. We ran our simulations on Ubuntu machines (version: 18.04) so all the commands are for Ubuntu.To run the simulations, you should follow the following steps:
- Step 1: Install dependencies
- Step 2: Install Omnet++
- Step 3: Clone the repository
- Step 4: Build the project
- Step 5: Run the simulations and extract the results
To successfully install Omnet++ and run the simulations, execute the following commands:
sudo apt update
sudo apt-get install -y build-essential
sudo apt-get install -y flex bison
sudo apt-get install -y zlib1g-dev
sudo apt-get install -y libxml2-dev
sudo apt install -y sqlite3
sudo apt-get install -y libsqlite3-dev
sudo apt-get install -y zip unzip
Make sure you have python3 installed (if not, run sudo apt-get install -y python3). If it is installed and accessible in "/usr/bin/python3" run the following:
sudo ln -s /usr/bin/python3 /usr/bin/python
sudo apt install -y python3-numpy
sudo apt-get install -y python3-matplotlib
The complete instructions for installing Omnet++ can be found in Omnet++ installation guide. In case you already have the simulator installed, you can skip this step. To install Omnet++ on an Ubuntu OS, you should run the following commands:
wget https://github.com/omnetpp/omnetpp/releases/download/omnetpp-5.6.2/omnetpp-5.6.2-src-linux.tgz
tar xvfz omnetpp-5.6.2-src-linux.tgz
cd omnetpp-5.6.2/
. setenv
For the next line we are assuming that you have extracted omnet in $HOME. If this is not true, replace $HOME with the path you have extracted the omnet files to.
echo "export PATH=$HOME/omnetpp-5.6.2/bin:\$PATH" >> ~/.bashrc
source ~/.bashrc
./configure WITH_QTENV=no WITH_OSG=no WITH_OSGEARTH=no
In case, the configure command printed out that "omnetpp-5.6.2/bin" is not added to your path, close and re-open your terminal and run the configure command again. If you are not using GUI, reboot your system using sudo reboot. Make sure you get the "Your PATH contains /opt/omnetpp-5.6.2/bin. Good!" message in the configuration process before moving forward.
make
To clone the repository, run the following script:
git clone https://github.com/hopnets/vertigo_simulations.git
We have provided a build.sh shell script to simplify this process. To build the project modules, run the following scripts:
cd vertigo_simulations/Omnet_Sims/
./build.sh
Considering that running the simulations included in the paper takes a long time (about 2-3 weeks) we provide you with two sets of configurations:
- A configuration file for running small scale simulations using 1Gbps links instead of 10Gbps and 40Gbps links. In this setup we create 25% background load and change the load generated by incast flows from 10% to 70%.
- The configuration files required for running heavy simulations that are included in the paper.
We have provided bash scripts for running both these sets and extracting their results. We have also included a sample python file which is used to process the extracted results of the small scale simulations and compute the Query Completion Times (QCTs) and Flow Completion Times (FCTs). To run the small scale simulations, extract their results, and perform the measurements, run the following commands:
cd dc_simulations/simulations/Vertigo_Sims
bash download_dist_files.sh
bash run_simple_1g.sh
The commands above, download the distribution files for the small scale simulations and simulate the following scenarios:
- DCTCP + ECMP
- DCTCP + DRILL
- DCTCP + DIBS
- DCTCP + Vertigo
Every scenario is expected to take less than 3 hours. After the simulations are over, our bash script automatically runs the python code, print the results, plot the figures, and save them in the ./figs directory. The sample output printed for the case with 95% load should be as following:
4,spines,8,aggs,40,servers,1,burstyapps,1,mice,40,reqPerBurst,19.75,bgintermult,0.071,burstyintermult,250,ttl,2,rndfwfactor,2,rndbouncefactor,20000,inc
astfsize,0.00120,mrktimer,0.00120,ordtimer,0,rep,dctcp,ecmp.csv
type,requested, started, finished, %Flow Completion, mean, p10, p20, p50, p90, p99, p99.9
all,700629, 644206, 500509, 71.43709438233358, 0.9035198559196383, 0.0027676920115998985, 0.23800162310879955, 0.7700354720689999, 2.0077613670308008,
3.4086337941725975, 4.282782992595809
mice of size, 3272, 3189, 3180, 97.1882640586797, 0.0779965320272783, 5.651716879997438e-05, 0.00016592004219999622, 0.0006479080000001414, 0.00134734
32311002839, 1.0011150451317201, 2.627085622420876
cat of size, 3407, 3330, 1886, 55.35661872615204, 0.6408765407683458, 0.001762701393499988, 0.008222333916000024, 0.209886923125, 2.1534168505725, 3.68
0864207729554, 4.371477864332964
elephant of size, 25, 23, 4, 16.0, 0.14661606853300013, 0.10581463790990003, 0.1121618918198001, 0.12065566903300029, 0.20818581875610012, 0.2419355787
3500995, 0.24531055473290103
bursty,693925, 637664, 495439, 71.39662067226286, 0.9098244404951409, 0.0029989001767999567, 0.2646591219934001, 0.7704694512530001, 2.014488231308199
8, 3.40962390337562, 4.286127312187507
background,6704, 6542, 5070, 75.62649164677804, 0.2874375921124215, 9.315383320003258e-05, 0.00039761385199992287, 0.0009834576329996736, 1.00056002410
12, 3.2408673568643787, 4.039009506724958
Queries, 0, 17500, 1186, 6.777142857142857, 3.443625160091087, 2.6232832229145, 2.89414442277, 3.4466851690614995, 4.2760554903334995, 4.7126774274068,
4.9133858131497306
4,spines,8,aggs,40,servers,1,burstyapps,1,mice,40,reqPerBurst,19.75,bgintermult,0.071,burstyintermult,250,ttl,2,rndfwfactor,2,rndbouncefactor,20000,inc
astfsize,0.00120,mrktimer,0.00120,ordtimer,0,rep,dctcp,drill.csv
type,requested, started, finished, %Flow Completion, mean, p10, p20, p50, p90, p99, p99.9
all,703832, 641659, 477687, 67.86946316734675, 1.0298296357981376, 0.0036188928988001314, 0.3491261024022001, 0.876681708508, 2.161909185377001, 3.5131
20593832041, 4.335185638332323
mice of size, 3286, 3208, 3201, 97.41326841144249, 0.05495433622796219, 5.853442999992353e-05, 0.00016172419599991272, 0.0006790667139999762, 0.001320
0741100005686, 1.0010064390370002, 2.5613873491084593
cat of size, 3434, 3359, 1779, 51.80547466511357, 0.5805412308059658, 0.001609507964200141, 0.0039674866104000735, 0.16146413850600005, 1.8509151254294
003, 3.6145161040695, 4.115616411179996
elephant of size, 26, 26, 3, 11.538461538461538, 0.10031560799166661, 0.09431294124460002, 0.09553894644520003, 0.09921696204700003, 0.1067577331165998
6, 0.10845440660725983, 0.10862407395632583
bursty,697086, 635066, 472704, 67.81143216188534, 1.0381279545102466, 0.15582109748029982, 0.3939309235071998, 0.87731663874, 2.1664716674509004, 3.51
572875198854, 4.338031374559926
background,6746, 6593, 4983, 73.8659946635043, 0.24262344505187536, 9.783677840004003e-05, 0.0003963509455999659, 0.0009710518150001235, 0.877555242361
4003, 3.0386230910765604, 3.9416558355236884
Queries, 0, 17500, 1172, 6.6971428571428575, 3.492059415592647, 2.5950040902034996, 2.901149741557, 3.548579388255, 4.3844759362044, 4.815061341139679,
4.932485087877952
4,spines,8,aggs,40,servers,1,burstyapps,1,mice,40,reqPerBurst,19.75,bgintermult,0.071,burstyintermult,250,ttl,2,rndfwfactor,2,rndbouncefactor,20000,incastfsize,0.00120,mrktimer,0.00120,ordtimer,0,rep,dctcp,dibs.csv
type,requested, started, finished, %Flow Completion, mean, p10, p20, p50, p90, p99, p99.9
all,706137, 624386, 411080, 58.2153321522594, 1.0646161792927584, 0.001512368817000032, 0.17786951161260012, 0.8170995053414996, 2.4407175085992, 3.803333011258371, 4.529922768871086
mice of size, 3291, 3240, 3226, 98.02491643877241, 0.05101097443430658, 5.901520450002362e-05, 0.0008176479999999764, 0.0013882537564999797, 0.0023305600000000926, 1.00149849237175, 2.6297263568915756
cat of size, 3439, 3220, 1121, 32.59668508287293, 0.6864049231728181, 0.001441562464000068, 0.0027685607480001373, 0.03286976585500012, 2.370972504787, 3.938498857517, 4.764982883554406
elephant of size, 26, 23, 3, 11.538461538461538, 0.1281168424240001, 0.09983762160320006, 0.10082622967340012, 0.10379205388400026, 0.16612597866080003, 0.18015111173557996, 0.181553625043058
bursty,699381, 617903, 406730, 58.155711979593384, 1.073704947534576, 0.0016717374977000489, 0.2029768177814001, 0.8639440373825, 2.4460125040351013, 3.8057035188610926, 4.529962355688849
background,6756, 6483, 4350, 64.38721136767317, 0.21480590182277562, 9.787215050001699e-05, 0.000991770092799904, 0.0015462912245001181, 0.8771840705405998, 3.226393756749811, 4.4459843776404275
Queries, 0, 17500, 249, 1.4228571428571428, 3.5943158852147263, 2.5490543232758, 2.9075970346794, 3.736700805329, 4.6175067445566, 4.8243148405168, 4.926462310074928
4,spines,8,aggs,40,servers,1,burstyapps,1,mice,40,reqPerBurst,19.75,bgintermult,0.071,burstyintermult,250,ttl,2,rndfwfactor,2,rndbouncefactor,20000,incastfsize,0.00120,mrktimer,0.00120,ordtimer,0,rep,dctcp,vertigo.csv
type,requested, started, finished, %Flow Completion, mean, p10, p20, p50, p90, p99, p99.9
all,706760, 594809, 526221, 74.45540211670156, 0.8788331791513427, 0.674342980587, 0.676776130346, 0.7723981464309999, 1.465001677253, 2.6333061684715995, 3.3946427234425602
mice of size, 3294, 3293, 3293, 99.96964177292045, 0.00035123916248010423, 5.622321540017519e-05, 8.739019179988538e-05, 0.0001052160000001301, 0.00016321909359986455, 0.000545883972520001, 0.001543906542167682
cat of size, 3440, 1247, 841, 24.447674418604652, 0.3158612196790214, 0.0008680073169999858, 0.0015985511440002043, 0.018333252799, 0.900304798678, 3.4314982568872017, 4.277152951497738
elephant of size, 26, 6, 3, 11.538461538461538, 0.10142027584833337, 0.096515492409, 0.09732527268999999, 0.09975461353299997, 0.10699132421380012, 0.10861958411698014, 0.10878241010729815
bursty,700000, 590263, 522084, 74.58342857142857, 0.885285460173189, 0.6744910055508998, 0.6769474992097999, 0.772482567058, 1.4650339578776002, 2.6333134055875806, 3.3946122967514745
background,6760, 4546, 4137, 61.19822485207101, 0.06456373631608628, 7.340362440007853e-05, 9.136001039991016e-05, 0.00011375337600005864, 0.019338946428200204, 2.0580358969355044, 4.008590732094909
Queries, 0, 17500, 8005, 45.74285714285714, 1.6087287174550076, 0.7711473605348, 0.7731677268893999, 0.8809048311120002, 2.6367102924524, 3.3964733290294, 3.5918592925148682
The results above illustrate that, under 95% load, ECMP, DRILL, and DIBS complete 6.77%, 6.69%, and 1.42% of the querries, respectively, while Vertigo completes 45.74%.
Viewing the figures
To view the plotted figures for small simulations, Jupyter notebook is required. To install Jupyter using pip simply issue:
pip install jupyterlab. The notebook file is located in Omnet_Sims/dc_simulations/simulations/Vertigo_Sims/. To view the notebook in your browser, run:
jupyter Omnet_Sims/dc_simulations/simulations/Vertigo_Sims/result_illustrator.ipynb
You can optionally use --ip x.x.x.x flag on a public server machine, where x.x.x.x is the public ip address of the server.
The notebook compares the small scale simulation results with the large-scale results reported in the main paper.
Running large-scale simulations
The config files for large scale simulations can be used for evaluating Vertigo, DIBS, ECMP, and DRILL while using TCP, DCTCP, and Swift as the transport protocol. Every scenario with these configurations takes 2 to 3 weeks to complete. To run the large scale simulations, first make sure that you are in the right directory ("vertigo_simulations/Omnet_Sims/dc_simulations/simulations/Vertigo_Sims") and then run the following command to download the distribution files:
bash download_dist_files.sh
After the distribution files are downloaded, you can use the provided bash scripts to run the large scale simulations for different incast arrival rates (dqps), flow sizes, and scales. Additionally, you can run the simulations for various degrees of burstiness, fat-tree topology, and component analysis. The list of the provided bash scripts is as below:
- Different arrival rates with 15%, 25%, 50%, and 75% background load
- run_15_bg_dqps.sh (uses
omnetpp_15_bg_dqps.ini) - run_25_bg_dqps.sh (uses
omnetpp_25_bg_dqps.ini) - run_50_bg_dqps.sh (uses
omnetpp_50_bg_dqps.ini) - run_75_bg_dqps.sh (uses
omnetpp_75_bg_dqps.ini)
- run_15_bg_dqps.sh (uses
- Different scales with 50% background load
- run_50_bg_dscale.sh (uses
omnetpp_50_bg_dscale.ini)
- run_50_bg_dscale.sh (uses
- Different flow size with 50% background load
- run_50_bg_dfsize.sh (uses
omnetpp_50_bg_dfsize.ini)
- run_50_bg_dfsize.sh (uses
- Different degrees of burstiness with 80% constant load
- run_80_constant_dburstiness.sh (uses
omnetpp_80_constant_dburstiness.ini)
- run_80_constant_dburstiness.sh (uses
- Fat-tree topology under different arrival rates and 50% background load
- run_fattree.sh (uses
omnetpp_fattree.ini)
- run_fattree.sh (uses
- Component analysis: effect of boosting factor in Vertigo's performance (boosting factor = 0, 2, 4, 8)
- run_boosting.sh (uses
boosting.ini)
- run_boosting.sh (uses
- Component analysis: effect of SRPT scheduling, packet deflection, and ordering component in Vertigo's performance
- run_deflection_and_ordering.sh (uses
deflection_and_ordering.ini)
- run_deflection_and_ordering.sh (uses
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent