Archipelago: New Cloud Storage Backend of GRNET

Archipelago:New Cloud Storage Backend of GRNET

Filippos Giannakosphilipgian@grnet.gr

Chrysostomos Nanakoscnanakos@grnet.gr

Greek Research and Technology Network (GRNET) S.A.

16th TF-Storage MeetingFebruary 13, 2015

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 1 / 21

Background & Motivation Background

Background

Synnefo

OpenStack compatible cloud platform

Powers:

~okeanoshttps://okeanos.grnet.gr

~okeanos-globalhttps://okeanos-global.grnet.gr

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 2 / 21

Background & Motivation Motivation

Motivation

Pithos (Storage Sevice)

Swift APIUploaded Files / Images

Plankton (Image Service)

Glance APIRegistered Images

Cyclades (Compute/Network/Volume Service)

Nova/Neutron/Cinder APIVirtual Disks / Volumes

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 3 / 21

Background & Motivation Motivation

Unified View of Storage Resources

FilesUser files, with Dropbox-like syncing.

ImagesTemplates for VM creation.

VolumesLive disks, as seen from VMs.

SnapshotsPoint in time snapshots of volumes.

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 4 / 21

Archipelago Features Storage Features

Archipelago

Storage Virtualization System

Powering storage in Synnefo

Decouples storage resources from storage backends

Files / Images / Volumes / Snapshots

Unified way to provision, handle, and present resources

Decouples logic from actual physical storage

Software-Defined Storage

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 5 / 21

Archipelago Features Storage Features

Backend Agnostic

Advantages:

No vendor lock-inMigration from one storage backend to anotherPossibility of combining backends

~okeanos started with NFS and migrated to RADOS

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 6 / 21

Archipelago Features Storage Features

Storage Backend for the Cloud

With Archipelago we offer:

DeduplicationSyncable File/Image upload with partial uploadImage RegistrationThin clonesDownloadable thin snapshots

with zero data movement.

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 7 / 21

Archipelago Features Services Overview

Services Overview

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 8 / 21

Under the Hood Everything is a Resource

How it is Done

Everything is a resource on Archipelago

The same resource is exposed as:

A file through the API of the Storage Service (Pithos)An image through the API of the Image ServiceA live disk / VM volume through the API of the Volume ServiceA snapshot through the API of the Volume Service

All data remain in one place

No copying of data around

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 9 / 21

Under the Hood Mapfiles

Mapfiles

Mapfile for each resource

Keeps track of the mappings from resource offset to objectsKeeps metadata information for each resource

Operates on mapfiles

not on the actual data

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 10 / 21

Under the Hood Mapfiles

From Mapfiles to Data

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 11 / 21

Under the Hood Architecture

High Level Architecture

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 12 / 21

Under the Hood Architecture

Low level Architecture

Same Archipelago software stack on each node

Modular/Pluggable architecture

Distinct processes

Communicate via requests on a shared memory segment

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 13 / 21

Under the Hood Architecture

Low level Architecture

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 14 / 21

Under the Hood Architecture

Core

Volume composer

Composes resource I/O from individual objects

Mapper

Keeps and updates the mappings from resource offsets to individualobjects which actually hold the data

Flexible I/O pipeline

Can be extended with other components offering common functionality

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 15 / 21

Under the Hood Interfaces

Client Endpoints

Upstream Native QEMU virtio driver

VM interface

Blktap2 driver

Admin/VM interface

Pithos

Web interface / REST API

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 16 / 21

Under the Hood Interfaces

Backend Drivers

Currently supported backends:

Files (Any FS with POSIX semantics)RADOS (Using native librados interface)GlusterFS (Using native libgfapi interface)S3/SWIFT (Experimental)

General: Anything that can support object semantics

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 17 / 21

Advantages over other Solutions

Advantages over other Solutions

Allows for resource composition from already existing objects

Standalone resources

Unified view of all cloud resources

Backend agnostic

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 18 / 21

Experience

Experience from Production

3000 VM volumes

≈2.8M user uploaded files

≈20TB deduplicated data

≈80TB total data before replication

22M actual objects on the storage backend

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 19 / 21

The Future

Upcoming Features

New mapfile format

More compactAllows for larger volumes with minimal metadata overhead(1PB → 2GB)

Garbage collection

Deferred reference countingAutomatic deletion of unused objects

SSD Read-only caching

Lower latency

ISCSI target

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 20 / 21

Thanks

Thanks

Software lives at:

https://www.synnefo.org

Repository:

https://github.com/grnet/archipelago

Documentation:

https://www.synnefo.org/docs/archipelago/latest

Mailing list:

synnefo@googlegroups.com

F. Giannakos, C. Nanakos Archipelago 16th TF-Storage Meeting 21 / 21