Transcript: #StandardsGoals for 2024: What’s new for BISAC - Tech Forum 2024
Run tier 1 apps in the cloud with dedicated storage
1. How to run tier 1 applications in the cloud
Cloud Expo Europe 2013
Slides @ http://www.slideshare.net/SolidFireInc
Mark Thomas - Solutions Architect
Dave Wright – Founder / CEO
@jungledave
2. Introduction
Mark Thomas, Solutions Architect, Databarracks
IT Consultant for 14 years
Working with virtualisation since 2002 (ESX 1.5)
Working with cloud services since 2009
Worked for VirtualizeIT for 10 years – Consultant / Manager / Director
One of first VAC partners in UK
Production systems primarily from day one
Worked for Virtustream for 3 ½ years – Manager and Director
One of first specific cloud providers formed for that purpose
Gartner Magic Quadrant for IaaS
Dave Wright, Founder / CEO, SolidFire
• Left Stanford University in 1998 to help start GameSpy Industries
• GameSpy merged with IGN Entertainment as Chief Architect
• Founded Jungle Disk in 2007, sold to Rackspace in 2008
• Founded SolidFire in 2009 after experiencing the difficulties of high-
performance cloud storage
3. About Databarracks
A decade of experience in cloud computing
True managed services
Focussed on security and support
Certified
Backup to disaster recovery to Infrastructure as a Service
Cloud Integrators
Cloud Industry Forum Members
5. Our customers
Norway
Russia
Canada UK
Germany
Poland
SWZ
USA
Spain Italy Greece
China
UAE India
Hong Kong
Thailand
Singapore
Brazil
Peru
Australia
RSA
7. Common issues for tier 1 applications
Vendor / ISV support
Vendors say they don’t support virtual /cloud systems
Providers struggle to understand the complexities of tier 1 applications in these environments
Perception that performance is worse in a consolidated / multi-tenant
environment
Assumptions that hypervisors add an overhead and don’t perform as well as physical
Assumes all systems are many-to-one
Lack of understanding of cloud about resource SLA’s
Storage is an ‘unknown’ at provider and is key
WAN impact on performance
Moving systems away from local introduces latency to sensitive systems
Cost and reliability of links
It’s not secure enough
View that multi-tenant clouds are not secure enough to host tier 1 data
8. The noisy neighbour
A VM or VMs consuming all the resource
As total of all VM resources grows above the
total physical capacity. Additional resources
are required to satisfy demand or adjusting
VM setting to compensate. Performance
issues potentially but easy to manage BEYOND CAPACITY
RESOURCES CONSUMED
100%
50%
x
x
x
x
100% 100% 100% 100%
VM operate within configured resources. Actual
physical resource allocated by virtualisation
administrator
0% 0% 0% 0% BASELINE CAPACITY
CLIENT A
OVER TIME
9. The noisy neighbour: worse in multi-tenancy?
Same as before but now with multiple customers, whose VMs get adjusted?
Service provider managed maximum physical
resources. Service provider dictates
potentially how conflict situations are
resolved. Not always visible to clients
BEYOND CAPACITY
RESOURCES CONSUMED
100%
50%
x
x
x
x
100% 100% 100% 100%
VMs operate within configured
resources. Actual physical resource
allocated by service provider service
and clients commitment
0% 0% 0% 0% BASELINE CAPACITY
CLIENT A CLIENT B
OVER TIME
11. Compute ring-fencing
Reserving compute
Ensure application profiling identifies resource requirements and sensitivity
Treat as you would a physical system / your own virtualisation
Ensure your service provider understands your environment and has associated services to
support your migration
Dedicating physical compute
For things that don’t fit the provider’s shared platform
Not as efficient, pay for all physical resources
Customer owned or fully managed?
12. Storage auto tiering
Hides IO issues with SSD / fast disk for hot blocks
Hard to control who uses what and charge back
Client Volumes
Physical SAN
Disk pool with
two or more Standard Tier
Fast Tier
types of disk.
Normally less in Physical Disks
fast Tier
Blocks move A B C
A B C Blocks
as accessed A B C
A B C
13. Storage disk reservation
Dedicating disk to provide write IO performance
Cost of IO / capacity is a problem
Not really cloud storage
Clients A-C Client D
Clients A-C
share a set of Client Volumes
disks. Client D
has dedicated
Physical SAN
Both disk pools
have the same
number and
type of disks. Physical Disks
Same IO but
different usage
14. Re-engineer for the cloud
Businesses should evaluate the benefits of doing things differently
Applications that can be web enabled should be
Consider re-developing applications using cloud platforms for scalability and performance
Running applications from remote data centres is not new, architect the same way
Look at changing data access to optimise the traffic (move all services together, move client
desktops close to servers)
Leverage good service provider’s knowledge
Should have good SLA’s for resources
Should be able to understand Enterprise applications not just web scale
Should have options for optimising your environment (resource reallocation, tiers)
Should offer complimentary services such as WAN optimization / remote desktops etc. to
support the re-architecture
16. How we address the IO Issue
Storage for tier 1 applications can be placed on
SolidFire storage for IO SLA’s
Each client has their own logical volume
separation for performance monitoring and
security segregation
Each client has their own dedicated datastores to
leverage hypervisor level IO optimisations if
required
17. Customer IO visibility for tier 1
Price per capacity (included IOPS per GB), IO increase (priced per IO)
Customers can see volume IO setting via the Databarracks Service Portal
Customers can specify IOP requirements in the portal
18. The Noisy
Neighbor Effect Noisy Neighbor
• Few resource hungry
applications
negatively affect all
Traditional Multi-Tenant
other volume Performance
performance
19. Performance
The Cloud
• Unable to manage performance independent of
Needs Better capacity
Storage • Can not guarantee storage performance
Efficiency
• Low and inefficient utilization rates
• Lack of high performance in-line data reduction
Management
• Complex, manual, lacks automation
Scale
• Limited scalability of both capacity and
performance
• Manage multiple islands of storage
20. Key Differentiators
Guaranteed Quality In-line Complete Cloud
of Service (QoS) efficiency automation Scalability
In-line data reduction and REST-based API for Simultaneous scaling of both
Fine-grain performance
85% utilization requires less complete control capacity and performance
management on a per volume
purchased capacity
basis
21. SolidFire QoS
in Practice
Traditional Multi-Tenant
Performance
• Applications allocated Application of SolidFire
into performance tiers QoS controls
• Changes are immediate
can be made on the fly
22. Key • All-SSD Architecture
Requirements • Quality of Service functionality that
guarantees application storage
for running all performance
Tier 1 apps in • Firm performance based SLAs
the cloud.
A decade of experience in cloud computingBegan in 2003 providing online backup to consumers and to businessTrue managed servicesSince our inception, even when the business was just two directors, we have always had 24/7 support and that attitude has stuck with us in everything we doFocussed on security and supportWe host these services from the UK’s most secure data centres and we have a big team of engineers whose job is not just to fire-fight and fix problems, but to actually proactively monitor our customers and solve issues before they turn into problems. CertifiedOne of the first to become certified to the Cloud Industry Forum Code of Practice. ISO 27001 for information security and going through PCI/DSS and IL3 for G-Cloud and government data. Backup to disaster recovery to Infrastructure as a ServiceOur story as to how we got to where we are today is very organic. We started as a company providing online backup and recovery services. In a disaster we would bring that data to the customer’s DR site and restore it and get them up and running. Some customers wanted faster restores, so we introduced a Virtual Disaster Recovery service, restoring their servers in our data centres. Somewhere along the line, people starting referring to these services as “cloud” services. We then started to get a lot of enquiries from those VDR customers about doing this more permanently and so we have now moved into Infrastructure as a Service. Cloud IntegratorsIn parallel to this – customers also started to ask us to help them transition into other cloud and which is why we are also an AWS consulting partner
It’s the age old problem since server consolidation started with virtualisation. How do you ensure that the resource are used by the servers that need them at the right time. Lots of advancements in the hypervisors to combat this (reservations, shares resolutions, storage and network control). While this has enabled the forward movement of the virtualisation and cloud sector this is leading to more emphasis on the hypervisor vendors as the key component. Hardware follows suit to provide better performance to off load this function (CPU virtualisation extensions etc) ultimately changing the hardware landscape. While these technologies work well for single clients installs where its is very easy to decide who should get what resource due to one point of ownership
Cloud multi-tenancy introduces many more variables which many software and hardware platforms were just not ready for. Service providers now provider consolidated resources for 10s, 100s, 100s clients and has to decide how this resource management works. Is it manual is it automated are the technologies suitable?