CloudOpen Europe 2014
Avoiding the 1000 Dollar VM in your first cloud
#whoami
Name: Tim Mackey
Current roles: XenServer Community Manager and Evangelist; occasional coder
Cool things I’ve done • Designed laser communication systems
• Early designer of retail self-checkout machines
• Embedded special relativity algorithms into industrial control system
Find me • Twitter: @XenServerArmy
• SlideShare: slideshare.net/TimMackey
Clouds are cool
Agility • Ease of provisioning w/o hardware latency
• Remove IT as a service delivery critical path
• Manage complete infrastructure, regardless of
scale
Lower operational expenses • Consistent application and service deployment
• Drive reduced capital requirements
• Visibility into user and line of business usage
Why we want a cloud – the reality
Agility • Ease of provisioning w/o hardware latency
• Remove IT as a service delivery critical path
• Manage complete infrastructure, regardless of
scale
Lower operational expenses • Consistent application and service deployment
• Drive reduced capital requirements
• Visibility into user and line of business usage
Our boss bought the hype!!!
Why we want a cloud – the reality
Agility • Ease of provisioning w/o hardware latency
• Remove IT as a service delivery critical path
• Manage complete infrastructure, regardless of
scale
Lower operational expenses • Consistent application and service deployment
• Drive reduced capital requirements
• Visibility into user and line of business usage
Our boss bought the hype …
… but doesn’t want to look like an idiot
Today’s realities versus tomorrow’s norms
IT is a large capital expense
Work only happens in the office
IT does support and maintenance
Calling the helpdesk
IT is a monthly operating expense
Work happens wherever you are
IT drives strategic initiatives
On-demand, self-service IT
IT Today IT Tomorrow
Enterprise datacenter Cloud
• Architected for 100s of hosts
• Scale-up (server clusters)
• Applications assume reliability
• IT Management-centric
• Proprietary vendor stack
• Architected for 1000s of hosts
• Scale-out (multi-site server farms)
• Applications assume failure
• Autonomic [1:1,000’s]
• Open, value-added stack
Competing paradigms
Service Offerings
Clearly define what you want to offer • What types of applications
• Who has access, and who owns them
• What type of access
Define how templates need to be managed • Operating system support
• Patching requirements
Define expectations around compliance and availability • Who owns backup and monitoring
Define tenancy requirements
Department data local to department • Where is the application data stored
Data and service isolation • VM migration and host HA
• Network services
Encryption of PII/PCI • Where do keys live when data location unknown
• Need encryption designed for the cloud
Showback to stakeholders • More than just usage, compliance and audits
Let’s build a cloud ….
Where to start …
Deliver something easy • Web services are a perfect start
• With experience grow the install base
Pay for what you need • Use free hypervisors
• Use free orchestration services
Succeed early • Remember your success is tied to happy boss
Enterprise system assumptions
Redundant networking • LACP or bonding
• Stacked switches
• Additional NICs
High performance storage • iSCSI/ Fiber Channel
• Multiple paths
High density systems management • Blade servers
• Data center wide operations managment
Let’s get some shiny new gear …. … because you can’t just reuse what you’ve got
Redundant, independent networks • Management, storage, VM traffic 6 NICs
High performance storage • 10 Gbps links for storage and VM traffic
• 1 Gbps for management traffic
Compute simplification • Blade based servers
• Network fabric for network management
VM specification • 2 vCPU, 8GB RAM, 40GB shared disk
The BOM
Total hardware cost: 293 242 USD
Component Cost
HP BladeSystem Enclosure w/Flex 10 Interconnects 70 210 USD
Compute blades (14x BL460c, dual socket, 128GB) 157 536 USD
Cisco Nexus switches (2x5548UP) 36 620 USD
HP MSA Storage (24 drives/RAID 5/900GB) 28 876 USD
1222 USD per VM 3086 USD for equivalent physical server
Did you build a cloud?
Cost • Did we just refresh some hardware?
• Have we improved anything?
• Is this just “business as usual”?
Capabilities • Are we more agile?
• Can we take advantage of new paradigms?
• Is this easier to manage?
Scalability and failure
Fact • Cloud providers have outages
• Networks go down
• Usage varies
Assertion • Is failure a form of scalability?
• Can we leverage this in our design?
• Is our paradigm correct?
Attack the storage paradigm
Shared storage growth and provisioning time
1,000
500
VMs
Cost, AU 100 200
500
VMs
Provisioning efficiency
AU – arbitrary units
Combined efficiency and storage evolution
Redesign
1,000
500
VMs
100 200 Cost, AU
VMs
1,000
500
Cost, AU 100 200
? Alternatives
AU – arbitrary units
Redesign
Efficiency and pod storage
1,000
500
VMs
100 200 Cost, AU
POD #1
POD #2
POD #3 1,000
500
VMs
100 200 Cost, AU
AU – arbitrary units
No redesign
What about local storage?
1,000
500
VMs
Cost, AU 100 200
50
VMs
Provisioning efficiency
AU – arbitrary units
POD
trend
Traditional
trend
Cost-Performance Trends
Shared Storage Local Storage
1,000
500
VMs
Cost, AU 100 200
1,000
500
VMs
100 200 Cost, AU
Local storage
Performance
trend
Local storage
trend
Understanding relationship between VM density and IO
Plan B
The power of local storage
Our IO requirements • 300-400 IOPs per blade
• 3-4 SAS 10K disks, or SSD MLC
Our storage requirements • 700-900 GB per blade
Options • Storage blade (per pod)
• Local storage per blade
• Change to pizza boxes
Let’s get some shiny new gear …. … because you can’t just reuse what you’ve got
Redundant, independent networks • Management, VM traffic 4 NICs
No shared storage • 10 Gbps links for VM traffic
• 1 Gbps for management traffic
Compute • Rack based servers
VM specification • 2 vCPU, 8GB RAM, 40GB shared disk
The BOM
Total hardware cost: 127 490 USD
Component Cost
Dell R630 (dual socket, 128GB, 4x300GB SAS 10k) 98 880 USD
Cisco Nexus switches (2x3064X) 28 610 USD
531 USD per VM 56% savings over “enterprise best practices”
83% savings over physical server
Understanding what you want to accomplish
Public clouds are people carriers and minibuses
YOUR cloud should be a race car • Optimize it for your needs
Don’t rent what you can own cheaper • Cloud operator doesn’t care about your success
• Optimized applications might be key
Ensure you have backup plans • Usage can and does spike
• Outages can and do happen
vs.
Virtualization infrastructure choices
Hypervisor defined by service offerings • Don’t select hypervisor based on “standards”
• Understand true costs of virtualization
• Multiple hypervisors are “OK”
• Bare metal can be a hypervisor
To “Pool” resources or not • Is there a real requirement for pooled resources
• Can the cloud management solution do better?
• Real cost of shared storage
Primary storage defined by hypervisor
The ROI of a private cloud OpEx vs CapEx
EC2 costs (US/East) • m3.large (Linux) on-demand: 1352 USD per year
• m3.large (Linux) 3yr reserved: 1609 USD for three years
• m3.large (RHEL) 3 yr reserved: 3348 USD for three years
Your cloud costs • 115 130 USD depreciated over 3 years: 183 USD per VM per year
• Operational costs vary
Become a service provider to your business • Provide on demand services, which scale with the business
• Through business understanding, retain operational control of IT
• Simplify compliance monitoring
Tying it all Together
1. Define success criteria
2. Evaluate current best practices
3. Select a topology which works
4. Decide on storage options
5. Define supported configurations
6. Select preferred hypervisor(s)
7. Build your Cloud
WORK BETTER. LIVE BETTER.