29
Hyperconverged Cloud Stack with Synnefo, Ganeti and Ceph Thomas Schend, Principal Systems Architect, CompuGroup Medical AG Ceph Day Berlin 28. April 5/7/2015 Ceph Day Berlin 1

Ceph Day Berlin: Hyperconverged cloud stack with Synnefo, Ganeti and Ceph

Embed Size (px)

Citation preview

Hyperconverged Cloud Stack with

Synnefo, Ganeti and CephThomas Schend, Principal Systems Architect, CompuGroup Medical AG

Ceph Day Berlin

28. April

5/7/2015 Ceph Day Berlin1

Agenda

5/7/2015 Ceph Day Berlin2

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

Bio

5/7/2015 Ceph Day Berlin3

CGM AG, eHealth Software Company

Systems Architect, responsible for Infrastructure Design

OSS Enthusiast

LPIC-2,MCITP,VCP,CCA,DCIE,…

Contact: [email protected]

Agenda

5/7/2015 Ceph Day Berlin4

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

What do we need from IaaS?

5/7/2015 Ceph Day Berlin5

Reliable

Scalable (Backend and VMs)

Easy to manage

Long lived workloads, not all VMs are EC2 instances ;-)

Synnefo

5/7/2015 Ceph Day Berlin6

Open Source

Production-quality IaaS

made for Persistent VMs

commodity hardware only

Large-scale Installations

simple UI for the end-user

uses Ganeti as a Cluster Manager

Small project

Synnefo History

5/7/2015 Ceph Day Berlin7

Design started late 2010

Production since July 2011

~okeanos* public cloud running currently ~8500 VMs

*http://okeanos.grnet.gr

Synnefo Layers

5/7/2015 Ceph Day Berlin8

Image provided by GRNET

Synnefo Architecture

5/7/20159

Image provided by GRNET

Ceph Day Berlin

Synnefo II

5/7/201510

Build with Ganeti as a backend

Astakos: Identity/Account services

Pithos: File/Object Storage service

Cyclades: Compute/Network/Image/Volume services

snf-manage commandline tool

helpdesk tool

snf-burin / snf-burin-run

kamaki commandline client

Code on github

Ceph Day Berlin

Agenda

5/7/2015 Ceph Day Berlin11

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

Ganeti History

5/7/201512

Very mature, production ready

Open Source since August 30, 2007 by google

uses Mature open source tools LVM,DRBD,python,haskell

Different hypervisors (Xen,KVM,LXC)

Low overhead for cluster management

Ceph Day Berlin

Scale from 1-200 nodes and beyond

Live migration without shared storage

Noderoles: master,master candidate,regular nodes

Nodegroups

OVF import tool

Backup with included tool

Job queue

Commands starts with gnt-

Ganeti features

5/7/201513 Ceph Day Berlin

No need for special hardware (SAN)

But SAN/Shared Storage is supported

Manage cluster as an entity not single nodes (libvirt)

Few dependencies

No database

Easy recovery in case of hardware outage

No SPOF which can affect the reliability of VMs

Very good documentation

Ganeti advantages

5/7/201514 Ceph Day Berlin

Deployment automation puppet,chef,cfengine,..

Monitoring check_mk/icinga/nagios,…

Open vSwitch LACP + 10GBe

Autorepair

Packages for all major OS debian,ubuntu,rhel/centos

Ganeti best practices

5/7/201515 Ceph Day Berlin

Agenda

5/7/2015 Ceph Day Berlin16

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

Running ceph giant

Heavy use of cache pool + SSD journals (write-back)

Mon + osd + kvm in same nodes

Custom crush map for flash + spinners in same nodes

Tweaked recovery parameters

Ceph facts

5/7/201517 Ceph Day Berlin

Archieplago overview

5/7/201518 Ceph Day Berlin

Archieplago interfaces

5/7/201519 Ceph Day Berlin

Archieplago

5/7/201520 Ceph Day Berlin

Archieplago resource composition

5/7/201521 Ceph Day Berlin

Archieplago processes

5/7/201522 Ceph Day Berlin

Agenda

5/7/2015 Ceph Day Berlin23

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

2U x86 Server

2x CPUs 10 Core Intel

300+ GB RAM

8 10k SAS Drives

2 OS Drives (RAID1)

2 400GB PCIe SSDs

Dual Port 10GBe

OOB Management

Hardware

5/7/201524 Ceph Day Berlin

Agenda

5/7/2015 Ceph Day Berlin25

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

Isolation of resources (Compute vs Storage)

Looking into svirt, cgroups, docker, etc.

Maintenance also affects Storage availability / migration

Needs proper processes for operations / noout

Scaling Compute / Storage ratio

Needs different node “types”

Custom CRUSH MAP integration into deployment

puppet

Challenges

5/7/201526 Ceph Day Berlin

Agenda

5/7/2015 Ceph Day Berlin27

Introduction

Synnefo

Ganeti

Ceph with Archipelago

Hardware

Challenges

Q & A

http://www.synnefo.org

https://github.com/grnet/synnefo

https://code.google.com/p/ganeti/

Try it!

5/7/201528 Ceph Day Berlin

5/7/201529

Q & A

Ceph Day Berlin