Eipifi / SwiftCloud

Swiftcloud

SwiftCloud: a causally-consistent CRDT object database for client-side apps

This repository contains the prototype database SwiftCloud, originally described in the Inria tech report "SwiftCloud: Fault-Tolerant Geo-Replication Integrated all the Way to the Client Machine" by Marek Zawirski, Annette Bieniusa, Valter Balegas, Sérgio Duarte, Carlos Baquero, Marc Shapiro, and Nuno Preguiça (http://hal.inria.fr/hal-00870225), with an updated description appearing in a PhD thesis of Marek Zawirski on "Dependable Eventual Consistency with Replicated Data Types" (http://pagesperso-systeme.lip6.fr/Marek.Zawirski/papers/thesis.pdf). This codebase served in the experiments for the latter.

The code is licensed under The Apache License Version 2.0 (LICENSE), provided in the spirit of CRAPL license (CRAPL-LICENSE), i.e., it is ugly to read/use and far from production ready.

SwiftCloud crash course

TODO

Example applications

• SwiftSocial - a simple social network application modeled after WaltSocial
• Port of YCSB benchmark
• File system with FUSE bindings

Running scripts

Deployment scripts for experiments are implemented in Groovy. The scripts are in scripts/groovy/deployment directory, both deployment classes definitions (such as SwiftBase.groovy, SwiftSocial2.groovy etc.) and concrete experiment instances (runsocialmanual.groovy etc.).

To run the scripts, you need:

• groovy and ant installed on your machine, and
• bin/ directory with compiled sources (e.g., generated by Eclipse by default)
• ssh access configured for the target machine. In the case of EC2 machines used in experiments, the simplest is to configure your ssh like this:

Host *.compute.amazonaws.com
	StrictHostKeyChecking no
	UserKnownHostsFile=/dev/null
	IdentityFile PATH_TO_SSH_KEY_GOES_HERE

Host *.compute-1.amazonaws.com
	StrictHostKeyChecking no
	UserKnownHostsFile=/dev/null
	IdentityFile PATH_TO_SSH_KEY_GOES_HERE

To run experiments, use scripts from scripts/groovy/swift/deployment E.g. to run a single swiftsocial experiment manually:

./scripts/groovy/swift/deployment/runsocialmanual.groovy

(TODO: edit that example to modify SwiftSocial2 parameters)

E.g. to run a series of experiments oriented at scalability:

./scripts/groovy/swift/deployment/runscalabilitythroughput.groovy

The experiments generate log files. To generate plots from the logs, use scripts from eval_R/. e.g. to generate a detailed analysis of a run:

Rscript eval_R/analyze_run.R all <run_logs.tar.gz> [output directory]

e.g. to compare different runs use eval_R/compare_runs.R and associated Makefiles.