QUISP
The Quantum Internet Simulation Package (QuISP) is an event-driven
simulation of quantum repeater networks, which will be the foundation
of the coming Quantum Internet. QuISP's goal is to simulate a full
Quantum Internet consisting of up to 100 networks of up to 100 nodes
each. Its focus is on protocol design and emergent behavior of
complex, heterogeneous networks at large scale, while keeping the
physical layer as realistic as possible.
QuISP is a product of the Advancing Quantum Architecture (AQUA) research group headed by Prof. Rodney Van Meter, at Keio University's Shonan Fujisawa Campus, Fujisawa, Japan. See http://aqua.sfc.wide.ad.jp and the list of Authors.
Research questions
A simulator is one or more of three things: a time machine, an X-ray machine, or a telescope. It can be used to look into the future or the past, look at the internals of an object that are otherwise inaccessible, or look at objects of a far vaster scale than can be built in a laboratory. QuISP is, perhaps, most importantly a telescope: we are looking at large-scale quantum networks and ultimately the Quantum Internet, but we do also pause to look at the micro (local) behavior of protocols, as well.
Research questions we hope to answer:
- Emergent behavior
- Classical networks exhibit congestion collapse; are quantum networks subject to the same thing?
- Will the dynamics of large networks prevent us from making end-to-end connections under realistic scenarios, even when a naive model suggests it should be possible?
- Are there other unexpected behaviors in large-scale networks?
- Protocol design
- Testing of detailed protocol designs to validate correct operation.
- Are there interactions between the classical and quantum portions of the network?
- Connection architecture and performance prediction
- All three proposed network generations exhibit complex behavior that makes analytic prediction of performance difficult with realistic parameters; simulation, of course, will require the best effort we can make at validation, as well.
- Dynamic behavior
- Are networks stable as conditions change?
- When a topology change occurs, how do the protocols respond?
- Network traffic will be dynamic; can our multiplexing and resource management protocols adapt so that new connections are given service in a reasonable time, and ongoing connections continue to receive expected levels?
Simulation goals
We have a number of long-term goals for development of the simulator:
- Complex network topologies, including the notion of network boundaries and heterogeneity at the physical and logical levels
- support 1G, 2G and 3G quantum networks, utilizing either purify-and-swap (1G) or quantum error corrected (QEC) (2G and 3G) protocols for managing errors
- Distinct link architectures: memory-to-memory (MM), midpoint interference (MIM), and midpoint source (MSM), sneakernet, satellite
- Internetworking protocols for connecting different types of networks
- Various applications running in complex traffic patterns
Because these protocols can result in hundreds of qubits in a single entangled state, and the entire system may consist of up to a million qubits, simulation at the physical Hamiltonian level or even just above that at the unitary (gate, e.g. CNOT) level is infeasible. We cannot calculate and store full density matrices for such states. Instead, like simulators for large-scale error correction, QuISP operates primarily in the error basis, in which we maintain a description of errors the states have incurred rather than the full state. However, unlike most QEC simulators, QuISP supports non-Pauli errors, in a limited fashion.
QuISP is almost endlessly configurable; for example, it is possible to set not only different lengths for different links in the network, but also different gate error rates and memory lifetimes on individual qubits. Non-Pauli errors that are at least partially supported in the current release include qubit loss, relaxation to ground state, excitation to excited state, and complete mixing.
If you are unfamiliar with the research literature or the terminology above, see "Learning more", below.
In addition, we aim to make simulations run on QuISP completely
reproducible, to the extent humanly possible. It will be possible
for others to verify work done using QuISP if they have the name of
the QuISP release, version numbers for supporting software, the .ini
file, any changed .ned
files, and the seed for the pseudo-random
number generator.
Current status
Most of the infrastructure is up and running, though the sets of actual experiments (interesting simulations) you can do are fairly limited still.
Working infrastructure:
- All of the basic OMNeT++ functionality, such as events, visualization, logging, analysis, etc.
- Complex network topologies, defining per link parameters including length, channel error rates, numbers of qubit memories, gate error rates, etc. (You will see the included demonstration networks in the documentation linked below.)
- A complete internal model of the software architecture for a repeater, including a connection manager, the RuleSet execution engine, real-time tracking of quantum states, etc.
Besides the obvious joys of the endless network configurability, here are the key quantum protocols that are implemented:
- basics of RuleSet creation & distribution
- various purification protocols: Single round of X purification, alternating X/Z purification, etc. Default to one round of purification per entanglement swapping round. Extending these to test your own custom purification protocol is pretty straightforward.
- tomography: when the simulation boots, it assumes that the software at each end of each link knows nothing about the link, so it begins by performing tomography on the links. (This is actually problematic, because it turns out to take a long time for tomography to converge, which means a lot of boot-up time in the simulation before other interesting things start to happen. We are working on a way to pre-calculate this, so that you can choose to either include tomography or not.)
- entanglement swapping.
Current generic networking-level status:
- fully blocking circuit switching
- random pairwise traffic pattern (flat distribution)
In-progress work:
- entanglement swapping is a relatively new feature, and taking data using it is still not fully implemented
- although the connection setup protocol works, the teardown on completion of a connection still needs a little love
- the set of demo networks is still being polished
Upcoming features in near-term releases:
- more general resource allocation & multiplexing
- more general mechanism for establishing traffic patterns
- MSM links
- graph states at the link level
- multi-party states at the application level
Mid-term to long-term release features:
- 2G networks, esp. Jiang style
- full quantum internetworking
Installation requirements
The full installation process is described in Wiki. The main software tools you will need are:
- QUISP requires OMNeT++ and
- an external C++ library, Eigen, to work.
- To contribute to QuISP development, you will also need to be familiar with at least the basics of git.
- We recommend the use of Doxygen for source code comments, but the Doxygen tools are not required unless you want to build the source code documentation.
Depending on your local setup and how you intend to use QuISP, you may also need various tools (a C++ compiler, make, an X Windows server, Docker, ffmpeg for making videos, etc.), documented in the installation notes.
Trying it out on the web
If you just want to take a peek at the basic sample simulation set, we encourage you to try them out on the web (here) which is built using the WebAssembly. Currently the Wasm version only supports running the pre-configured simulations and users cannot upload custom topology, we are still working on that. Also due to the heavy load of the OMNeT++ and the QuISP itself the performance on the web version is a lot slower than running it locally.
Building and running locally
First see Wiki, then follow the instructions below.
There are two main ways of working with QUISP. You can either use the
Eclipse-like graphical interface of OmNET++, for which you will find
instructions in Building QuISP with OMNeT IDE
,
or you can use the Makefile
and GNU make, by looking at instructions
in Building QuISP with GNU Make
(wiki). Some operations are
implemented in the Makefile and not explained for the graphical user
interface.
Moving into useful work
Once you have gotten this far, you should be able to run some of the most basic demos. Next, you'll want to learn how to create your own test networks, and how to extend the source code for your own uses.
When you are ready to start contributing, you can start reading the code, as we have done.
You will also want to look at some of the software design documents.
Development tools
A few tools (mainly scripts) can be used to make development a bit easier.
Look around in the bin
folder of this project.
Is QuISP right for me?
Fundamentally, the point of QuISP is that networks are much more than point-to-point connections.
If you want to know about the behavior of systems and networks, to study behavior of links that are too complex for simple analytic equations (esp. those with multiple qubits per link) or to contribute to the design of network protocols, QuISP is the simulator for you. If you're trying to adjust detector window timing v. entanglement fidelity, or figure out what Q factor your cavity needs to be, or understand dispersion in a fiber, it might not be.
Learning more
See the references.
Contributing
First, join the QuISP Slack team.
Please also refer to the code of conduct and Contributing guide.
License
QuISP was initially released on April 5, 2020, and is licensed under the 3-Clause BSD License.
QuISP builds on OMNeT++. OMNeT++ itself is a custom license, open source and free for academic use, but a license fee required for commercial organizations. QuISP also requires the linear algebra library Eigen, where license is MPL2, and so probably not an issue.