Self-service virtual environments at Deutsche Telekom based on - - PowerPoint PPT Presentation
Self-service virtual environments at Deutsche Telekom based on - - PowerPoint PPT Presentation
Self-service virtual environments at Deutsche Telekom based on OpenStack Alexander Stellwag Deutsche Telekom AG Products & Innovation Agenda Introduction Short History Of Wolke-7 Planning and Setup Virtual Private
Agenda
- Introduction
- Short History Of Wolke-7
- Planning and Setup
- Virtual Private Environments
- Version 2: havana and neutron
- Lessons Learned
- Q&A
Introduction
- Alex Stellwag
- 43 Years
- Started at Deutsche Telekom
in 2001
- Cloud Architect since 2011
- OpenStack since „Diablo“
Short History of Wolke-7
- Started back in late 2011 after major reorganization to gain knowledge of
OpenStack
- Dec 2012: initiated two basic use cases
–
Plain IaaS, configurable via EC2-API
–
Fully managed environments (PaaS)
- Both available from one cloud environment
- Internal customers only, no public services
- Core principles
–
Stay open source
–
Build over partnering
Intranet Infrastructure
- Supermicro servers (2UTwin² and 2U 12 Disk)
- Ubuntu 12.04 LTS (Kernel 3.2 and 3.11)
- OpenStack packages from UCA
- Ceph packages from Inktank
- I/S automation via FAI1 and puppet
- Monitoring via nagios / check_mk
1 http://fai-project.org
Some Figures
- 12 compute nodes with 2x Xeon X5650 (6 cores, 2.67 GHz,
Intel HT enabled), 96 GB Ram, 2x 250 GB SSD each
- 3 Ceph nodes with 2x Xeon E5-2630, 64 GB Ram, 2x 240 GB
SSD + 10x 2TB S-ATA (OSDs) each. ==> ~60 TB net
- 2x Cisco Nexus 5548 for tenant VLANs and floating IPs
- 2x Cisco 48 port 1 GE for administration
Planning and Setup
- Setting up OpenStack is quite painless thanks to the official docs
- We needed a complete DC
– Location – Connectivity – Basic I/S services (DNS, Mail, Proxy) – Automation services (FAI, puppet)
- Complex Setup due to internal security requirements
- Adapting the puppet modules for our needs took a while
Virtual Private Environments I
- Simple applications (apache, mysql, tomcat)
– Provided by user-data – Published on project website – User has to copy & paste them into dashboard – Designed to fulfill security requirements – Full self-service via OpenStack dashboard – Users have full control and responsibility
#cloud-config # this cloud config code installs apache2-worker webserver # and changes several configuration directives to already meet # some GIS requirements # after customization of the webserver the requirements have to be # checked agein # supported OS: Ubuntu 12.04 # addtional packages to be installed packages:
- apache2
- libapache-mod-security
# commands to modify the default installation runcmd:
- a2dismod cgid # req 7
- sed -i 's/\(.*Alias \/\(cgi-bin\|doc\)\)/#\1/; //{s/^/#/}; //,/Options /{s/\(.*\n\t\t\tdeny from all\n\t\t \
<\/LimitExcept>/}' /etc/apache2/sites-available/default # req 9,11,12,13,16,18
- sed -i 's/^/#/' /etc/apache2/mods-available/alias.conf # req 16
- sed -i 's/^\s*ServerTokens .*/ServerTokens Prod/' /etc/apache2/conf.d/security # req 19
- service apache2 restart # restart webserver to apply changes
Virtual Private Environments II
Virtual Private Environments III
- Complex environments (currently CI/CD for Java only)
– Jenkins, git, tomcat – Set up via bash and python scripts – Not yet integrated in dashboard, ordered via e-mail – Fully managed by Test & Development team
Version 2: havana & neutron
- Started in 10/2013
- New location, so again we needed to setup a complete DC
- Ready for production in May 2014
- Currently looking for dedicated operations team
- Heat installed and enabled but Virtual Environments not ported
- ver yet.
- Users may use their own stack templates
Infrastructure Changes
- No dedicated network node, all agents distributed amongst
compute nodes
- Multiple L3 and DHCP agents per L3 network
- Custom python script to check availability of agents and
reschedule if necessary
- Works, but we have no long-time experience
- Currently looking into Icehouse, but migration has issues
(mainly OS-wise)
New Hardware
- 11 compute nodes with 2x Xeon E5-2660 v2 (10 cores, 2.2
GHz, Intel HT enabled), 128 GB Ram, 2x 400 GB SSD each
- 8 Ceph nodes with 2x Xeon E5-2620 v2 (6 cores, 2.1 GHz, Intel
HT enabled), 32 GB Ram, 2x 100 GB SSD + 10x 4TB S-ATA (OSDs) each. ==> ~312 TB net
- 2x Alcatel Lucent OS 6900-T40 for VxLAN tunnels and floating
IPs
- 2x Alcatel Lucent OS 6850 for administration
Lessons Learned
- I/S deployment is hard – even if you start on a green field
- Deploying OpenStack without the ability to fix the code is
difficult (and sometimes frustrating)
- HA for neutron is a complex beast
- Deploying multi-tier applications without heat is no fun
Lessons Learned II
- Manage internal stakeholder for proper funding
- Many customers don't know the difference between
virtualization and cloud computing
- Change management program needed because cloud impacts
95% of operations roles
- For large and inhomogeneous teams SCRUM is not the right