Skip to content

Latest commit

 

History

History
146 lines (100 loc) · 3.61 KB

README.md

File metadata and controls

146 lines (100 loc) · 3.61 KB
Raw

MCC (Massive Client Connections)

Introduction

MCC is a distributed network workload generator based on user-space stack. It can replicate flow-level network load utilizing BSD-like socket API provided by user-space TCP/IP stack (mTCP). According to experiments, MCC can generate 3.2 million concurrent TCP connections using only a single CPU core, and achieve almost linearly scalability in multicore platform and distributed systems.

Architecture

Prerequisites

Intel DPDK
libpcap libnet cmake3
boost-devel
bc
fmt pciutils

Included directories

./    
|__apps/			Sample applications      
|__conf/			Sample configuration files       
|__http/			HTTP parsing files      
|__mtcp/			User-level stack      
|__scripts/			Scripts used for deployment when running in distributed mode    
|__src/				Source files    
|__lib/				Static-Link libraries
|__tests/			Testing examples

Introduction to apps

  • mcc_client: Application used for simulating massive concurrent tcp connections in IoT scenerio.
  • http_loader: an HTTP benchmarking tool like wrk.
  • wan_loader: Application used for generating Web workload.

Installation

Downloads

$ ./install_dependencies.sh
$ git submodule init
$ git submodule update
$ cd mtcp
$ git submodule init
$ git submodule update

Setup User-level stack

(You may reference to https://github.com/mtcp-stack/mtcp for details)

$ cd mtcp/
$ ./build-dpdk.sh

Compile

$ cd ..
$ ./build.sh

There is a build_type option in build.sh to designate build type of the project. And the built executable file is put in directory $PWD/build/$build_type/

Run

  • Single Machine

Take mcc_client for example:

$ cd build/release/apps/mcc/
$ ./mcc -c 1000000 -d 60 --smp 9 --network-stack mtcp --dest 192.168.3.6

-c [ --conn ] arg (=100)		Total connections
-d [ --duration ] arg (=0)		Duration of test in seconds
--smp arg (=1)					Number of threads (default: one per CPU)
--network-stack arg (=kernel)	Select network stack (default: kernel stack)
--dest arg (=192.168.1.1)		Destination IP
  • Cluster Mode

Take distributed_mcc_client for example:

$ ./run-master.sh
$ ./run-workers.sh 64 start hosts
$ ./run-workers.sh 64 stop hosts

Scripts used above are in scripts/ directory.

Frequently asked questions

  1. How can I quit the application? Use ^C to gracefully shutdown the application. Or you can kill process with “pkill” command in Linux.

  2. No connection is established successfully when using user-level stack. You need to add static ARP entry and static route entry.

Release Notes

V 2.0

Add support for distributed mode and stateless workload generation

  • Distributed
  • Stateful/Stateless

V 1.2

Add suport for application-level load generation

  • Encryption/decryption
  • Simulating response processing
  • Modeling for packet length and inter-departure time

V 1.1

Add supprt for paket-level priority

  • MCC only supports flow-level priority before, which means packets in the same flow hold the same priority for MCC v1.0. To meet more needs, we add support for packet-level priority.

V 1.0

Multicore scalable network load generator based on userspace stack.

  • User-level-stack-based
  • Reactor pattern
  • Multi-threaded

Contacts

[email protected]

[email protected]

References

[1] Wu, Wenqing, et al. "MCC: a Predictable and Scalable Massive Client Load Generator." International Symposium on Benchmarking, Measuring and Optimization. Springer, Cham, 2019.