README.md 9.62 KB
Newer Older
1
# SimBricks
Jialin Li's avatar
Jialin Li committed
2

3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
[![Documentation Status](https://readthedocs.org/projects/simbricks/badge/?version=latest)](https://simbricks.readthedocs.io/en/latest/?badge=latest)
[![Docker Hub](https://img.shields.io/badge/docker-hub-brightgreen)](https://hub.docker.com/u/simbricks)
[![Chat on Slack](https://img.shields.io/badge/slack-Chat-brightgreen)](https://join.slack.com/t/simbricks/shared_invite/zt-16y96155y-xspnVcm18EUkbUHDcSVonA)
[![MIT License](https://img.shields.io/github/license/simbricks/simbricks)](https://github.com/simbricks/simbricks/blob/main/LICENSE.md)

## What is SimBricks?

SimBricks is a simulator framework aiming to enable true end-to-end simulation
of modern data center network systems, including multiple servers running all
full software stack with unmodified OS and applications, network topologies and
devices, as well as other off the shelf and custom hardware components. Instead
of designing a new simulator from scratch, SimBricks combines and connects
multiple existing simulators for different components into a simulated full
system. Our primary use-case for SimBricks is computer systems, networks, and
architecture research. Our [paper](https://arxiv.org/abs/2012.14219) provides a
more detailed discussion of technical details and use-cases.

Currently SimBricks includes the following simulators:

- [QEMU](https://www.qemu.org) (fast host simulator)
- [gem5](https://www.gem5.org/) (flexible and detailed host simulator)
- [ns-3](https://www.nsnam.org/) (flexible simulator for networks)
- [OMNeT++ INET](https://inet.omnetpp.org/) (flexible simulator for networks)
- [Intel Tofino SDK Simulator](https://www.intel.com/content/www/us/en/products/network-io/programmable-ethernet-switch/p4-suite/p4-studio.html)
  (closed-source vendor-provided simulator for Tofino P4 switches).
- [FEMU](https://github.com/ucare-uchicago/FEMU) (NVMe SSD simulator).
- [Verilator](https://www.veripool.org/verilator/) (Verilog RTL simulator)

## Quick Start

Depending on how you plan to use SimBricks, there are different ways to start
using it. The quickest way to get started just running SimBricks is with our
pre-built Docker container images. However, if you plan to make changes to
SimBricks, you will have to build SimBricks from source, either through Docker,
or on your local machine. Below these methods are listed in order of increasing
effort required.

**Please refer to
[our documentation](https://simbricks.readthedocs.io/en/latest/) for more
details.**

### Using Pre-Built Docker Images

**This is the quickest way to get started using SimBricks.**

We provide pre-built Docker images on
[Docker Hub](https://hub.docker.com/u/simbricks). These images allow you to
start using SimBricks without building it yourself or installing any
dependencies. This command will run an interactive shell in a new ephemeral
container (deleted after the shell exits):

```Shell
docker run --rm -it simbricks/simbricks /bin/bash
Antoine Kaufmann's avatar
Antoine Kaufmann committed
56
57
```

58
59
60
If you are running on a Linux system with KVM support enabled, we recommend
passing `/dev/kvm` into the container to drastically speed up some of the
simulators:
Jialin Li's avatar
Jialin Li committed
61

62
63
64
```Shell
docker run --rm -it --device /dev/kvm simbricks/simbricks /bin/bash
```
Antoine Kaufmann's avatar
Antoine Kaufmann committed
65

66
Now you are ready to run your first SimBricks simulation:
Jialin Li's avatar
Jialin Li committed
67

68
69
70
71
```Shell
root@fa76605e3628:/simbricks# cd experiments/
root@fa76605e3628:/simbricks/experiments# simbricks-run --verbose --force pyexps/qemu_i40e_pair.py
...
Jialin Li's avatar
Jialin Li committed
72
```
Hejing Li's avatar
readme  
Hejing Li committed
73

74
### Experimental: Interactive SimBricks Jupyter Labs
Antoine Kaufmann's avatar
Antoine Kaufmann committed
75

76
**This is still a work in progress.**
Antoine Kaufmann's avatar
Antoine Kaufmann committed
77

78
79
80
81
82
We are working on a more interactive introduction to SimBricks through Jupyter
Labs in [this repository](https://github.com/simbricks/labs). These also simply
require starting a pre-built docker container and then connecting to it from
your browser. After this you can follow the interactive steps to run SimBricks
simulation directly from your browser.
Antoine Kaufmann's avatar
Antoine Kaufmann committed
83

84
### Building Docker Images
Antoine Kaufmann's avatar
Antoine Kaufmann committed
85

86
87
88
89
If you prefer to build the Docker images locally you will need `git`, `make`, and
`docker build` installed on your system. Other dependencies should not be
required. Now you are ready to build the docker images (depending on your system
this will likely take 15-45 minutes):
Antoine Kaufmann's avatar
Antoine Kaufmann committed
90

91
92
93
94
```Shell
git clone https://github.com/simbricks/simbricks.git
cd simbricks
make docker-images
Antoine Kaufmann's avatar
Antoine Kaufmann committed
95
```
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194

This will build a number of Docker images and tag them locally, including the
main `simbricks/simbricks` image.

### Building in VSCode Dev Container

**We recommend this approach if you plan to modify or extend SimBricks.**

This repository is pre-configured with a
[Visual Studio Code Development Container](https://code.visualstudio.com/docs/remote/containers)
that includes all required dependencies for building and working on SimBricks.
If you have Docker setup and the vscode remote containers extension installed,
you can just open a freshly cloned simbricks repo in vscode and vscode will
display a prompt to re-open the folder in the container. The vscode terminals
will also automatically run any commands inside of the container.

To compile the core SimBricks components simply run `make` (with `-jN` to
use multiple cores). Note that by default, we do not build the Verilator
simulation as these take longer to compile (one to three minutes typically)
and also skip building the RDMA proxy as it depends on the specific RDMA NIC
libraries. These can be enabled by setting `ENABLE_VERILATOR=y ENABLE_RDMA=y`
on the `make` command-line or by creating `mk/local.mk` and inserting those
settings there.

The previous step only builds the simulators directly in the SimBricks
repository. You likely also want to build at least some of the external
simulators, such as gem5, QEMU, or ns-3. You can either build all external
simulators by running `make -jN external` (this could take multiple hours
depending on your machine), or build them individually by running e.g.
`make -jN sims/external/qemu/ready` (replace `qemu` with `gem5`, `ns-3`,
or `femu` as desired).

Next, to actually run simulations, you will also need to build the disk images
with `make -jN build-images` (note this requires QEMU to be built first).
This will build all our disk images, while `make build-images-min` will
only build the base disk image (but not the NOPaxos or Memcache images used for
some experiments). This step will again take 10 - 45 minutes depending on your
machine and whether KVM acceleration is available but only needs to be run
once (unless you want to modify the images).

Now you are ready to run simulations as with the pre-built docker images.

### Building From Source

Finally, it is of course possible to install the required dependencies
directly on your machine and then build and run SimBricks locally. Note that
you will need to install both the modest build dependencies for SimBricks but
also for the external simulators you need. We suggest you refer to the
[`docker/Dockerfile.buildenv`](docker/Dockerfile.buildenv) for the
authoritative list of required dependencies.

## Repository Structure

- `doc/`: Documentation (sphinx), automatically deployed on
  [Read The Docs](https://simbricks.readthedocs.io/en/latest/?badge=latest).
- `lib/simbricks/`: Libraries implementing SimBricks interfaces
  - `lib/simbricks/base`: Base protocol implementation responsible for
    connection setup, message transfer, and time synchronization between
    SimBricks component simulators.
  - `lib/simbricks/network`: Network protocol implementation carrying ethernet
    packets between network components, layers over the base protocol.
  - `lib/simbricks/pcie`: PCIe protocol implementation, roughly modelling PCIe
    at the transaction level, interconnecting hosts with PCIe device simulators,
    layers over base protocol.
  - `lib/simbricks/nicbm`: Helper C++ library for implementing behavioral
    (high-level) NIC simulation models, offers similar abstractions as device
    models in other simulators such as gem-5.
  - `lib/simbricks/nicif`: *(deprecated)* Thin C library for NIC simulators
    establishing a network and a PCIe connection.
- `dist/`: Proxies for distributed SimBricks simulation running on multiple
  physical hosts.
  - `dist/sockets/`: Proxy transporting messages over regular TCP sockets.
  - `dist/rdma/`: RDMA SimBricks proxy (not compiled by default).
- `sims/`: Component Simulators integrated into SimBricks.
  - `sims/external/`: Submodule pointers to repositories for existing external
    simulators (gem5, qemu, ns-3, femu).
  - `sims/nic/`: NIC simulators
    - `sims/nic/i40e_bm`: Behavioral NIC model for Intel X710 40G NIC.
    - `sims/nic/corundum`: RTL simulation with verilator of the
      [Corundum FPGA NIC](https://corundum.io/).
    - `sims/nic/corundum_bm`: Simple behavioral Corundum NIC model.
    - `sims/nic/e1000_gem5`: E1000 NIC model extracted from gem5.
  - `sims/net/`: Network simulators
    - `sims/net/net_switch`: Simple behavioral Ethernet switch model.
    - `sims/net/wire`: Simple ethernet "wire" connecting two NICs back-to-back.
    - `sims/net/pktgen`: Packet generator.
    - `sims/net/tap`: Linux TAP device adapter.
    - `sims/net/tofino/`: Adapter for Intel Tofino Simulator.
    - `sims/net/menshen`: RTL simulation with verilator for the
      [Menshen RMT Pipeline](https://isolation.quest/).
- `experiments/`: Python Orchestration scripts for running simulations.
  - `experiments/simbricks/`: Orchestration framework implementation.
  - `experiments/run.py`: Main script for running a simulation.
  - `experiments/pyexps/`: Example simulation configurations.
- `images/`: Infrastructure to build disk images for hosts in SimBricks.
  - `images/kernel/`: Slimmed down Linux kernel to reduce simulation time.
  - `images/mqnic/`: Linux driver for Corundum NIC.
  - `images/scripts/`: Scripts for installing packages in disk images.
- `docker/`: Scripts for building SimBricks Docker images.