Cloud Storage [VLDB 25]

Prototype Code

This repository contains our prototype code implementing GlobalStore, a global object store built on top of the object stores in the cloud. It currently support S3, Azure Blob, and GCS. Users interact with SkyStore through S3 API.

This repository contains:

1. Multi-cloud traces and workloads
2. S3-Proxy: a fully compatible web server that speaks S3 protocol, communciates with Store-Server, and talks to Azure Blob, S3, and GCS underneath the hood
3. Store-Server: a FastAPI server that maintains namespace mappings and various policies. It speaks to the database (e.g. Postgres / SQLite) via SQLAlchemy model.

This repository is structured as follows:

• /prototype/experiment/trace - the IBM traces used in our benchmarks
• /prototype/s3-proxy/src - main implementation of S3-proxy written in Rust
• /prototype/store-server - the store-server implementations of virtual bucket and virtual object abstractions, along with placement and eviction policies

Prerequisites: see requirements.txt in /prototype/store-server

Test S3-Proxy and Store-Server Locally

To setup the environment:

• Ensure that you have Python and Rust toolchain installed.
• cargo install just. We use just as a task runner.

cd store-server
pip install -r requirements.txt

cd s3-proxy
just install-local-s3

# run the following commands in separate windows as they are blocking.
just run-skystore-server
just run-local-s3
just run

The S3 proxy should now be serving requests at http://localhost:8002.

You can use the AWS CLI or any S3 client to interact with the proxy. Do note that you will need to set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY to some dummy values. Checkout s3-proxy/justfile for reference.

Run some sample commands using the AWS CLI:

cd s3-proxy
just run-cli-create-bucket
just run-cli-list-buckets 
just run-cli-put
just run-cli-get
just run-cli-list
just run-cli-multipart

Test the server

cd store-server
just test

Setting Up Store-Server and S3-Proxy in remote VMs

• End-to-end benchmark is run in this script

Simulation code

This repository contains our simulation code implementing various policies in /simulation folder, the instructions is in README.md inside this folder.

The folder structured as:

• /simulation/SNIA_traces: scripts containing how we generate multi-cloud traces
• /simulation/benchmark: scripts on benchmarking real object stores
• /simulation/config: various policy config
• /simulation/src: poilcy implementation - our policy is named tevict_new and tevict_ranges_new

Usage

python main.py --config [CONFIG] --trace [TRACE] --vm [NUM_VMS] --setbase [SET_BASE] --simversion [SIM_VERSION]

2,3,6,9-region traces

The traces we use in our cost estimation is in /traces folder. Containing 2, 3, 6, 9 region, 5 SNIA IBM traces, with different types of workloads generated when number of regions is greater than 2.