highly available infrastructures based on xen · 2011. 2. 28. · - several implementations of xen...
TRANSCRIPT
Highly available infrastructures based on Xen Björn Brundert
2
Disclaimer
- All statements given in this presentation do not represent
the opinion of the speakers company but only the
personal opinion and experience of the speaker himself
3
Introduction, overview and scope
1) Virtualization and high availability
2) Hardware requirements and planning
3) Available solutions today
4) Future prospect
5) Summary
6) Q&A
1) Virtualization and high availability
- Google.com Define: high availability
- High availability is a system design protocol and associated
implementation that ensures a certain absolute degree of
operational continuity ...
en.wikipedia.org/wiki/High_availability
- But what is behind it?
1) Virtualization and high availability
- What does high availability (HA) mean?
- HA implies the definition of Service-Level-Agreements (SLA)
to measure e.g. the availability of a system
- SLAs represent the availability that a service provider plans
to keep
Availability per month Downtime per month
90% 72 hours
99% 7,2 hours
99,9% 43 minutes
99,99% 4 minutes
99,999% 25 seconds
6
1) Virtualization and high availability
- Standalone server does not meet HA-SLAs
- HA means redundancy clustering
- Cluster means higher complexity: building and operating
- Dedicated hardware per HA-service becomes more and
more expensive
- Hardware, infrastructure components, ETH/FC ports,
rackspace, power, cooling
- Active/passive cluster unused ressources
- How to achieve HA with virtualization
1) Virtualization and high availability
Application
Operating System
Hardware
Application
Operating System
HardwareXen-Hypervisor Xen-Hypervisor
Black Box Black Box01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
Shared Storage
8
1) Virtualization and high availability
- Virtualization offers enourmous benefits
- Consolidation: multiple services on one hardware
- Higher flexibility through VM encapsulation and hardware
independence
- Application does not require to be „cluster-ready“
9
2) Hardware requirements and planning
- Key rules when thinking about HA
- Be as paranoid as you want
- There is never enough redundancy
- Murphy is with you
- Identify single points-of-failure
- Consolidation means also higher impact in case of an
outage
- Example: Server
HP
ProLiant
DL380G5
UID 1 2
1 2 3 4 5 6 7 8
DIMMS
PCI
RISER
CAGE
FANS
PROCPROC
PP
M
PP
M
INTERLOCK
OVERTEMP
MIRROR
ONLINESPARE
POWER
SUPPLY
POWER
SUPPLY
10
2) Hardware requirements and planning
HP
ProLiant
DL380G5
UID 1 2
1 2 3 4 5 6 7 8
DIMMS
PCI
RISER
CAGE
FANS
PROCPROC
PP
M
PP
M
INTERLOCK
OVERTEMP
MIRROR
ONLINESPARE
POWER
SUPPLY
POWER
SUPPLY
11
2) Hardware requirements and planning
HP
ProLiant
DL380G5
UID 1 2
1 2 3 4 5 6 7 8
DIMMS
PCI
RISER
CAGE
FANS
PROCPROC
PP
M
PP
M
INTERLOCK
OVERTEMP
MIRROR
ONLINESPARE
POWER
SUPPLY
POWER
SUPPLY
HP
ProLiant
DL380G5
UID 1 2
1 2 3 4 5 6 7 8
DIMMS
PCI
RISER
CAGE
FANS
PROCPROC
PP
M
PP
M
INTERLOCK
OVERTEMP
MIRROR
ONLINESPARE
POWER
SUPPLY
POWER
SUPPLY
5 8
1 4BladeSystem
c3000
Enclosure
HP
OA2
OA1
Reset
Active
UID
NIC
1
NIC
2
UID
HP ProLiant
BL465c G6
NIC
1
NIC
2
UID
HP ProLiant
BL465c G6= ?
No!
12
2) Hardware requirements and planning
- I/O Cards:
NC340TFC1143 NC340TFC1143
FC NIC
13
2) Hardware requirements and planning
- Server
- Harddisk: RAID-configuration, spare
- Redundant I/O cards
- Redundant power supplies, redundant power cabeling and
redundant power circuits
- Redundant heartbeat-interconnect between nodes
14
2) Hardware requirements and planning
- Network components
- Redundant cabeling to server NICs
- Active/passive bonding needs to be set up in OS
- Redundant switch or redundant switch modules
15
2) Hardware requirements and planning
- Storage components:
- RAID configuration with hot-spare harddrives
- Redundant controllers
- Redundant cabeling to every node
- Terminating on different HBAs
- Storage based replication
16
3) Available solutions today
- Several implementations of Xen in the wild
- XenServer by Citrix
- Bundeled Hypervisor, out-of-the-box solution, HA-mechanism need to be bought with XenServer Essentials
- Version 5.5 released on June 16
- Free since Feburary 2009
- Xen in Linux-Distributions
- different implementations of HA-mechanisms like Red Hat Cluster Suite, heartbeat, pacemaker
17
4) Future prospect
- HA was only the first step
- Encapsulation is the key to more flexibility
- Fault-Tolerance is entering the market
- Xen Summit 2008: Demonstration of Kemari
- Marathon everRun VM: Available
- VMworld 2009 Europe: Announcement of VMware FT
18
4) Future prospect
- What is meant by FT?
- Virtual machine is running on two nodes simultaneously
- Primary machine is visible to the outside world
- Secondary machine is running as a shadow instance that
can take over any time no need for restarting the VM
Hardware Hardware
Xen-Hypervisor Xen-Hypervisor
VM 1 VM 2 VM 1 VM 2
19
4) Future prospect
- FT-mechanisms are on the roadmap for Xen 4.0
- Kemari (NTT, Japan): Unsupported version available
- Remus (UBC, Canada): Initial port to Xen unstable (3.4.0)
completed
20
5) Summary
- Virtualization brings more complexity to the infrastructure
- But: virtualization also brings more flexibility to the
infrastructure
- Obligatory redundant infrastructure for traditional cluster
systems can be „recycled“ and used for multiple OS
- High availability through encapsulation and hardware-
independence of OS as a virtual machine
- Fault-tolerance mechanisms are going to revolutionize
availability of stand-alone-servers
21
6) Q&A
- Any questions?