Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Cloudstack design decisions
Search
Pierre-Yves Ritschard
June 10, 2014
Technology
0
86
Cloudstack design decisions
Cloud operations at scale
Pierre-Yves Ritschard
June 10, 2014
Tweet
Share
More Decks by Pierre-Yves Ritschard
See All by Pierre-Yves Ritschard
Meetup Camptocamp: Exoscale SKS
pyr
0
440
The (long) road to Kubernetes
pyr
0
310
From vertical to horizontal: The challenges of scalability in the cloud
pyr
0
68
Change Management at Scale
pyr
0
110
5 years of Clojure
pyr
2
1k
Taming Jenkins
pyr
0
50
Init: then and now
pyr
1
200
Billing the Cloud
pyr
0
300
From Vertical to Horizontal
pyr
2
140
Other Decks in Technology
See All in Technology
AIのグローバルトレンド 2025 / ai global trend 2025
kyonmm
PRO
1
140
リモートワークで心掛けていること 〜AI活用編〜
naoki85
0
150
LLM 機能を支える Langfuse / ClickHouse のサーバレス化
yuu26
9
2.2k
全員が手を動かす組織へ - 生成AIが変えるTVerの開発現場 / everyone-codes-genai-transforms-tver-development
tohae
0
180
AIに目を奪われすぎて、周りの困っている人間が見えなくなっていませんか?
cap120
1
630
Claude Codeから我々が学ぶべきこと
oikon48
10
2.8k
家族の思い出を形にする 〜 1秒動画の生成を支えるインフラアーキテクチャ
ojima_h
3
1.1k
Lambda management with ecspresso and Terraform
ijin
2
160
20250807 Applied Engineer Open House
sakana_ai
PRO
2
380
結局QUICで通信は速くなるの?
kota_yata
3
3.2k
事業特性から逆算したインフラ設計
upsider_tech
0
110
大規模イベントに向けた ABEMA アーキテクチャの遍歴 ~ Platform Strategy 詳細解説 ~
nagapad
0
230
Featured
See All Featured
Optimizing for Happiness
mojombo
379
70k
Raft: Consensus for Rubyists
vanstee
140
7.1k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
358
30k
Designing for Performance
lara
610
69k
Statistics for Hackers
jakevdp
799
220k
Git: the NoSQL Database
bkeepers
PRO
431
65k
Documentation Writing (for coders)
carmenintech
73
5k
The Art of Delivering Value - GDevCon NA Keynote
reverentgeek
15
1.6k
How STYLIGHT went responsive
nonsquared
100
5.7k
The Art of Programming - Codeland 2020
erikaheidi
54
13k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
194
16k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
139
34k
Transcript
CLOUDSTACK DESIGN DECISIONS CLOUD OPERATIONS AT SCALE
SHORT BIO Pierre-Yves Ritschard CTO @ exoscale - The safe
home for your cloud applications Open Source Developer - pithos, cyanite, riemann, collectd, openbsd Architect of several cloud platforms - paper.li Recovering Operations Engineer
Simple and efficient cloud hosting platform Full compatibility with automation
tools Hosted in a safe jurisdiction
CLOUD BUILDING BLOCKS service infrastructure software people
SERVICE SIMPLICITY AND SCALABILITY Cloudstack based Basic networking Local storage
KVM hypervisor: SmartOS inspired
CLOUDSTACK Great extensibility, easy to plug into.
BASIC NETWORKING One IP per VM. Security groups are hypervisor
controlled layer 2 firewall rules. Provides all the flexibility of a traditionnal firewall, completely API controlled.
LOCAL STORAGE Fast I/O, persistent disks.
KVM HYPERVISOR Best in class hypervisor. Diskless and netboot approach.
Avoids resource waste, facilitates upgrades.
INFRASTRUCTURE THE GOOD CITIZEN CONTRACT Configuration management Visibility Build factory
Remote execution
THE GOOD CITIZEN CONTRACT new machines have roles role defines
converged configuration as sum of components each component has an expected normal state and reports it no local intervention needed
CONFIGURATION MANAGEMENT code is a great way to define infrastructure
ensures homogeneity ability to iterate fast great source of change tracking avoids fear of change
OVER 3000 COMMITS
CONFIGURATION MANAGEMENT: PUPPET battle tested tool simple declarative DSL to
express configuration fits our component approach well
VISIBILITY FROM THE MAP TO THE TERRITORY logs metrics alerts
WHY FOCUS ON VISIBILITY distributed systems with lots of moving
parts, high node volatility
LOGS all application and system logs sent over the wire
logstash disects and extracts metadata elasticsearch indexes for easy retrieval simple correlation
None
METRICS all application and system metrics sent over the wire
by collectd graphite's carbon aggregates and produces appropriate roll- ups if it moves, graph it. if it doesn't, graph it if it starts moving.
None
ALERTS unbounded stream of log and metric data passive approach
bodes well with node volatility riemann takes decisions based on stream content ability to extract meaningful information
BUILD FACTORY continuous integration package repositories
CONTINUOUS INTEGRATION over 60 build jobs ties into our code
hosting platform handled by jenkins
PACKAGE REPOSITORIES generates valid and signed Debian repositories ensures fast
upgrades simplifies configuration management
REMOTE EXECUTION a simple pubsub system recurrent commands stored as
scenarios command line, HTTP and IRC interaction
A SIMPLE PUBSUB SYSTEM each node runs an agent responsible
for carrying out commands. commands are sent to groups of nodes (by predicates such as role).
RECURRENT COMMANDS STORED AS SCENARIOS intricate workflows can be expressed
through a simple DSL
COMMAND LINE, HTTP AND IRC INTERACTION most of our production
environment can be controlled through our chatroom
SOFTWARE FILLING IN THE GAPS Customer management Real-time metering and
billing Integrated console A few other things
CUSTOMER MANAGEMENT Keeping track of our users Support services (ticket
management, coupons, emails)
REAL-TIME METERING AND BILLING can't be tied to a cloudstack
only solution cloudstack emits useful data ties into our customer management
INTEGRATED CONSOLE integrated experience across our services hides complexity and
cloudstack specifics exposes exoscale specific features
None
A FEW OTHER THINGS pithos cyanite fleet collectd add-ons
PEOPLE EFFICIENT WORK. QUIET NIGHTS Small SRE team Avoiding deploy
anxiety
SMALL SRE TEAM Our platform must be simple to operate,
additional moving parts must provide business value or help operations
AVOIDING DEPLOY ANXIETY Our software and infrastructure helps ensure we
have good tools to ensure quiet nights and easily caught errors
LOOKING BACK Cloudstack is a solid foundation for a IAAS
platform There's a bit more to it than just installing cloudstack Building a sustainable and scalable platform on top of cloudstack is possible
QUESTIONS ?