Skip to content

nislab/gensync-lib

BranchesTags

Repository files navigation

GenSync-lib

GenSync-lib is a library-generating wrapper around gensync-core, in a version suitable for packaging. This library provides a means for synchronizing data efficiently across different machines on MacOS and Linux. The library includes implmenetations of several state-of-the-art protocols that optimize different metrics, including:

  • Communication - total number of bytes transmitted
  • Rounds - the amount of back-and-forth needed to compelte synchronization
  • Computation - computing resources required
  • Memory - amount of physical memory needed

The current version is 2.1.1

Table Of Contents

Installation:

The easiest way to install the Gensync library libgensync.a is through mac ports:

port install gensync

Macports installs libgensync.a in the directory [MACPORTS dir]/lib, and the header files in [MACPORTS dir]/include. It should also install any needed dependencies.

Compilation from scratch

For non-apple machines, the library should be compiled from scratch, based on the following dependencies (for full-featured usage): * NTL - A library for doing Number Theory (>9.5) - ptheads - may be required depending on how NTL is configured - gmp - may be required depending on how NTL is configured * cppunit - For testing * cmake - For building

With dependencies installed, compilation proceeds as follows:

$ git clone --recurse-submodules [email protected]:nislab/gensync-lib.git
$ cd gensync-lib
$ cmake -B build
$ cmake --build build

This will produce the library libgensync.a in the build directory.

Example 1 - TryMe.cpp

This repository includes a file TryMe.cpp that demonstrates the Gensync library in action.

If the Gensync library and header files are placed within the default paths, the code can be compiled and run as follows:

$ g++ -std=c++11 TryMe.cpp -lgensync -lntl -o tryme
$ ./tryme

If the library and header files are placed in directories /opt/local/include and /opt/local/lib respectively (as typically done by mac ports), then the program can be compiled and run as follows:

$ g++ -I/opt/local/include -L/opt/local/lib -std=c++11 TryMe.cpp -lgensync -lntl -o tryme
$ ./tryme

On linux, the directories are typically /usr/local/include and /usr/loca/lib, but any directories can be subsituted above.

Output

This tryme program launches two processes, connected through a network socket:

  • The first process (host 1) contains a set with elements 'a', 'b', and 'c'.
  • The second process (host 2) contains a set of elements 'b' and 'd'.

The output from the program shows each host discovering the union of both sets (note that the order of elements within a set does not matter):

host 1 now has a b c d 
host 2 now has b d c a

Example 2 - TryMe2.cpp

A more generalized example can be found in TryMe2.cpp, which allows the user to select various synchronization parameters from the command-line.

It can be compiled into an executable tryme2 in a similar manner to tryme (just with TryMe2.cpp replacing TryMe.cpp).

To run the code, open two shell terminals.

Terminal 1

In one terminal, issue the command:

$ ./tryme2 server CPISync
connecting on port 8001...
sync succeeded.

This creates a server listening on port 8001 to synchronization requests from clients.

Terminal 2

In a second terminal, issue the command:

$ ./tryme2 client CPISync
listening on port 8001...
sync succeeded.

This connects

Extended Use Instructions:

  • Include Statements: #include <GenSync/(Aux/Data/Communicants/Syncs)/(The module you want to use)>

  1. Initialize a GenSync object with the constructor or builder helper class on the client and server machine

        GenSync::Builder builder = GenSync::Builder().
       setProtocol(GenSync::SyncProtocol::GenSync).  //GenSync,InterCPISync, ProbCPISync, IBLTSync, FullSync, etc.
       setComm(GenSync::SyncComm::socket). //CommSocket or CommString
       setPort(8001).
       setHost(localhost).
    
       //Set the parameters for your specific sync type (Not the same for every protocol. See GenSync build parameters section)
       setBits(64). // Bits per element
       setMbar(256). // Max differences to sync
       setErr(7); // -log_2(prob of error) to allow for errors that might otherwise underflow
       
    GenSync mySyncClientOrServer = builder.build();

  2. Add elements to your GenSyncs (If you need to add large elements use the ZZ class from NTL)

    • You may manually create a shared_ptr (Data/DataObject.h) or pass a data type compatible with DataObject and one will be automatically created for you, returning a pointer to the newly created DataObject
        mySyncClientOrServer.addElem(myZZ);
    mySyncClientOrServer.addElem(myLong);
    mySyncClientOrServer.addElem(myInt);
    mySyncClientOrServer.addElem(myMultiSet); // Multisets are serialized for use in IBLTSetOfSets. This does NOT add each element in the multiset to your sync.

  3. Run serverSyncBegin and clientSyncBegin on the server and client respectively

        mySyncServer.serverSyncBegin(0); //Add the index of the sync you would like to perform
    mySyncClient.clientSyncBegin(0);  //Multiple syncs or communicants may be added to one GenSync

  4. Collect relevant statistics

       //The sync index is decided by the order you added your sync in
   //If your GenSync has only 1 sync the index is 0
    mySyncClient.printStats(syncIndex); //Returns a string of formatted stats about your sync
    mySyncClient.mySyncVec[syncIndex]->getName(); //Retruns the name and parameters for your sync
    mySyncClient.getXmitBytes(syncIndex); //Returns the number of bytes transmitted by this sync
    mySyncClient.getRecvBytes(syncIndex); //Returns the number of bytes received by this sync
    mySyncClient.getCommTime(syncIndex); //Returns the amount of time in seconds that the sync spent sending and receiving info through a socket
    mySyncClient.getIdleTime(syncIndex); //The amount of time spent waiting for a connection or for a peer to finish computation
    mySyncClient.getCompTime(syncIndex); //The amount of time spent doing computations

GenSync Builder Parameters:

  • setProtocol: Set the protocol that your sync will execute (from the list above)
    • All syncs
  • setComm: Set the communication method your sync will use (CommSocket and CommString). Comm String is for local testing
    • All Syncs
  • setPort & setHost: Set the port & host that your socket will use
    • Any socket based syncs
  • setIOString: Set the string with which to synchronize
    • Only for CommString based syncs
  • setBits: The number of bits that represent each element in the set
    • All syncs except FullSync
  • setMbar: The maximum number of symmetric differences that can be synced by a GenSync
    • All GenSync variants
  • setErr: The negative log base 2 of the probability of error you would like to use to bound your sync
    • All GenSync variants
  • setHashes: If true, elements are hashed non-trivially (Must be true to synchronize multisets)
    • All GenSync variants
  • setNumPartitions: The number of partitions that InterCPISync should recurse into if it fails
    • InteractiveCPISync
  • setExpNumElems: The maximum number of differences that you expect to be placed into your IBLT. If you are doing IBLTSetOfSets this is the number of child sets you expect
    • IBLTSync, OneWayIBLTSync & IBLTSetOfSets
  • setExpNumElemChild: Set the upper bound for number of elements in each child set
    • IBLTSetOfSets
  • setDataFile: Set the data file containing the data you would like to populate your GenSync with
    • Any sync you'd like to do this with

Sync Types:

  • Included Sync Protocols (Sets and Multisets):
    • CPISync
      • Sync using the protocol described here. The maximum number of differences that can be reconciled must be bounded by setting mBar. The server does the necessary computations while the client waits, and returns the required values to the client
    • CPISync_OneLessRound
      • Perform CPISync with set elements represented in full in order to reduce the amount of rounds of communication by one (No hash inverse round of communication). The server does the necessary computations while the client waits, and returns the required values to the client
    • OneWayCPISync
      • Perform CPISync in one direction, only adding new elements from the client to the server. The client's elements are not updated. The server does the necessary computations and determines what elements they need to add to their set. The client does not receive a return message and does not have their elements updated
    • ProbCPISync
      • Perform CPISync with a given mBar but if the amount of differences is larger than that, double mBar until the sync is successful. The server does the necessary computations while the client waits, and returns the required values to the client
    • InteractiveCPISync
      • Perform CPISync but if there are more than mBar differences, divide the set into numPartitions subsets and attempt to CPISync again. This recurses until the sync is successful. The server does the necessary computations while the client waits, and returns the required values to the client
    • FullSync
      • The client sends the server its set contents and the server determines what elements it needs from the clients set. The server also determines what elements the client needs and sends them back.
    • IBLTSync
      • Each peer encodes their set into an Invertible Bloom Lookup Table with a size determined by NumExpElements and the client sends their IBLT to their per. The differences are determined by "subtracting" the IBLT's from each other and attempting to peel the resulting IBLT. The server peer then returns the elements that the client peer needs to update their set
    • OneWayIBLTSync
      • The client sends their IBLT to their server peer and the server determines what elements they need to add to their set. The client does not receive a return message and does not update their set
    • CuckooSync
      • Each peer encodes their set into a cuckoo filter. Peers exchange their cuckoo filters. Each host infers the elements that are not in its peer by looking them up in the peer's cuckoo filter. Any elements that are not found in the peer's cuckoo filter are sent to it.
  • Included Sync Protocols (Set of Sets):
    • IBLT Set of Sets
      • Sync using the protocol described here. This sync serializes an IBLT containing a child set into a bitstring where it is then treated as an element of a larger IBLT. Each host recovers the IBLT containing the serialized IBLTs and deserializes each one. A matching procedure is then used to determine which child sets should sync with each other and which elements they need. If this sync is two way this info is then sent back to the peer node. The number of differences in each child IBLT may not be larger than the total number of sets being synced

Reference:

If you use this software in your works, please cite the following paper (DOI):

Boškov, Novak, Ari Trachtenberg, and David Starobinski. "Gensync: A new framework for benchmarking and optimizing reconciliation of data." IEEE Transactions on Network and Service Management 19.4 (2022): 4408-4423.

or in BibTex:

@article{bovskov2022gensync,
  title={Gensync: A new framework for benchmarking and optimizing reconciliation of data},
  author={Bo{\v{s}}kov, Novak and Trachtenberg, Ari and Starobinski, David},
  journal={IEEE Transactions on Network and Service Management},
  volume={19},
  number={4},
  pages={4408--4423},
  year={2022},
  publisher={IEEE}
}

Additional literature

The code, explanation, references, API, and a demo can be found on this web page. If you use this work, please cite any relevant papers below.

The main theoretical bases for the approaches in this work are:

Relevant applications and extensions can be found at:

  • D. Starobinski, A. Trachtenberg and S. Agarwal, "Efficient PDA synchronization" IEEE Transactions on Mobile Computing 2:1, pp. 40-51 (2003). http://ipsit.bu.edu/documents/efficient_pda_web.pdf

  • S. Agarwal, V. Chauhan and A. Trachtenberg, "Bandwidth efficient string reconciliation using puzzles" IEEE Transactions on Parallel and Distributed Systems 17:11,pp. 1217-1225 (2006). http://ipsit.bu.edu/documents/puzzles_journal.pdf

  • M.G. Karpovsky, L.B. Levitin. and A. Trachtenberg, "Data verification and reconciliation with generalized error-control codes" IEEE Transactions on Information Theory 49:7, pp. 1788-1793 (2003).

  • More at http://people.bu.edu/trachten.

Additional algorithms:

  • Eppstein, David, et al. "What's the difference?: efficient set reconciliation without prior context." ACM SIGCOMM Computer Communication Review 41.4 (2011): 218-229.

  • Goodrich, Michael T., and Michael Mitzenmacher. "Invertible bloom lookup tables." 49th Annual Allerton Conference on Communication, Control, and Computing (Allerton), 2011.

  • Mitzenmacher, Michael, and Tom Morgan. "Reconciling graphs and sets of sets." Proceedings of the 37th ACM SIGMOD-SIGACT-SIGAI Symposium on Principles of Database Systems. ACM, 2018.

Contributors:

Elements of the GenSync project code have been worked on, at various points, by:

  • Ari Trachtenberg
  • Sachin Agarwal
  • Paul Varghese
  • Jiaxi Jin
  • Jie Meng
  • Alexander Smirnov
  • Eliezer Pearl
  • Sean Brandenburg
  • Zifan Wang
  • Novak Boškov
  • Xingyu Chen
  • Nathan Strahs

Acknowledgments:

About

No description, website, or topics provided.

Resources

Readme

License

GPL-3.0 license
Activity
Custom properties

Stars

0 stars

Watchers

2 watching

Forks

2 forks
Report repository

Releases 8

v2.1.0 Latest
May 6, 2024
+ 7 releases

Packages

No packages published

Contributors 10

Languages