Injustice - Developers Among Us (SciFiDevCon 2024)
Storage architectures and the cloud
1. Storage Architectures and the Cloud
& field notes from 18 months at SolidFire
Andy Roberts – andy.roberts@solidfire.com - Field Solutions
Architect
2. Career
Cloudwash
First day in storage was a Register worthy event
March 2002
50K Mailboxes lost with a storage system on fire
Supporting a SPaaS platform based on OpenSource
Customers bring VMWare into production
2004 – 2007 as VMWare becomes de facto
Everything very manual or with minimal scripting
First Clouds – 2007 - 2010
Early attempts at automation
Multi-tenancy becomes a daily focus
Cloud.Com appears on my radar - 2011
Cloud product evaluation
The birth of a Cloud
March 2013 – time for a change
First day at SolidFire – 8th April
Jully 2013 – first CloudStack meetup
Invested time into learning CloudStack & OpenStack
3. Public Cloud
Has Raised
the Bar
…driving the transformation of enterprise IT globally
4. The Silo’ed Infrastructure Model Is Unsustainable
Purchasing Marketing Sales Development
First Cloud NIH Cloud 3rd go… This will be
the one…
Source: http://www.wired.com/2012/05/stuck-in-silo/
5. Legacy Data Center
Single Tenant
Isolated work loads
Dedicated Infrastructure
Scale-up
Pre-provisioned capacity
Hardware defined
Project Based
Manual Administration
Next Generation Data Center
Multi-tenant
Mixed workloads
Shared infrastructure
Scale-out
Capacity on demand
Software defined
Self service
Automation
Entering a whole new world…
6. Regardless of the name, desired outcomes are the same
Next Generation Data Center
Software Defined
Data Center
Infrastructure 2.0 Cloud Computing
Private Cloud IT-as-a-Service
Agile Scalable Automated Predictable
7. A Fundamental Shift In App Design & Delivery
Legacy Data Center Next Generation Data Center
Apps span VMs
Data stored in scalable DBs,
cached in memory tiers
Interaction via API, web or mobile
Apps run on a computer
Data stored in files
Interaction via UI
8. CloudStack ?
OpenStack ?
Next Gen VMWare ?
All of the above?
Roll my own?
Commodity Stack?
Engineered Stack?
Leave the legacy behind?
Bring it with me?
Public or private?
9. The “Community” is what makes it work
Users
Service Providers
Vendors
Integrators
10. Expanding the Reach of The Next Generation Data Center
Agile
Scalable
Automated
Predictable
13. Scale-Out vs. Scale-Up
Performance
Capacity
Scale-Out
Performance
Capacity
Scale-Up
Limited performance scalability
Creates islands of storage with
poor utilization
Forklift upgrade model
Massive single system scale
Capacity and performance
managed as independent pools
Performance increases as you
grow capacity
New performance and capacity
resources available in minutes
14. Capacity
Performance
SF2405 nodes added
SF9010 nodes added
SF2405 Initial Cluster
Addition of future SF nodes
Mixed Node Clusters
Dynamically change capacity
and performance characteristics
of your cluster over time
15. Two views on benefits of scale
Service Provider & Internal Service Provider
Simplifies capacity management
Re-use across product lines as they come and go
Match spending to cash flow
Enterprise
Buy only what you need, rather than 3 years up-front
Continual Refresh of platform
Simplifies capacity management
16. Guaranteed Performance - QoS
Performance Virtualization: Independent
global pools of capacity and performance
Allocate: Storage performance
independent of capacity
Manage: Performance real-time without
impacting other volumes
Guarantee: Performance to every volume
with fine-grain QoS settings
Deliver Guaranteed Storage Performance and
Firm Performance SLAs
17. Two views on benefits of QoS
Service Provider & Internal Service Provider
Much better for multi-tenancy than just vLANs alone
Sell QoS as an advantage
Remove performance / noisy-neighbour headaches
Enterprise
The business will trust your XaaS platform
Match performance to LOB applications
Remove performance management headaches
18. Automated Management
Integrated REST-based API
Enables complete automation of any
SolidFire function
Supports development of user-facing
storage controls
Seamless integration into current
billing/charge-back systems and
management stacks
API Integration Support from SolidFire
19. Drastically Simplified Storage Management
SolidFire Architecture
Unified pool of capacity and performance
Automatic load distribution
Mixed-Node Cluster Support
Self-healing HA
Complete REST-based API automation
Traditional Storage Management
Separate pools of SATA, SAS, and Flash
RAID levels, aggregates, volume groups
Forklift controller and system upgrades
Fire drills on hardware failure
Manual UI Management
20. Views on benefits of Automation
Service Provider & Internal Service Provider
Just what the market expects
Enterprise
Reduce costs and/or drive out re-work
Increase agility and time to market
Enabler for all other benefits
QoS + Automation + Scale = Huge Compound Benefit
21. Personal View Storage marketplace is exciting as its ever been (!)
Great range of options
Blend multiple technologies together
Persistent or ephemeral
Cloud Automation Market
Amazing what you can achieve
Things really have come a long
Challenges as a vendor
Collaborate and differentiate at the same time
Give people the widest choice
Move up to the “Higher Ground”
23. We’ve been busy...
● Google Summer of Code
o http://www.solidfire.com/blog/mentoring-with-google-summer-
of-code-and-lessons-in-cloudstack/
o Automated Regression Testing
o Reworking host and storage tagging
o New GUI functionality for plugins
o Mentoring new talent as part of Apache Foundation
24.
25. CloudStack Storage Evolution
4.2 - dynamic provisioning for data drives
4.3 - adding KVM support
4.4 - root disk dynamic provisioning & resize
4.5 - GUI for storage plug-in & snapshot baby
steps
4.6 - planned : snapshot & clone integration
immutable UUIDs are a challenge
breaking the 256 LUN limit a focus
26. Whats next?
Come and see Mike present if you are at Collab
Catch-up on slides if not
Ideas appreciated on storage improvement
SolidFire Continue to Invest in CloudStack
andy.roberts@solidfire.com
( on behalf of mike.tutkowski@solidfire.com)
This is my career in storage.
I have cloudwashed this however I think its an interesting way of looking back at experiences, and how things are both changing and in many ways remain the same.
The fact is Public Cloud was the gateway to setting a new standard in how to run a Data Center.
The Public Clouds built their businesses on providing maximum efficiency and flexibility of their data center resources
This created a new benchmark for how IT is delivered.
The model of dedicated resources to specific groups in your organization was fine (10 years ago),
but today it equates to poor utilization of resources and increased management/administrative head-aches
How the same challenge is often true in service providers or companies that went to shared platform models early on
Silos of clouds remain and often it can be difficult to move between them.
This is what I think CloudStack, OpenStack & Open source clouds can help resolve, using community innovation and standard APIs
We’re moving from a world of isolated users, stifled by isolation and manual administration
To one of shared resource pools available in real time using self service tools
There’s a lot of different buzzword’s that are used to describe all of this.
The bottom line is regardless of what the “name” what we’re talking about is the “Next Generation Data Center”
We need agility, scalability and automation, ideally on predictable platforms that deliver on their promises
And while application development and delivery is changing (the devops model)
Fundamentally there is a huge opportunity in moving traditional and existing applications into more efficient infrastructures
So when I meet customers both in the service provider space, and those enterprises that want to act like service providers
I see people asking themselves, vendors and those around them the big directional questions.
Enterprises that see the benefits inherent in the Amazon-style approach are particularly asking themselves questions here.
If we then think about CloudStack & Cloud Automation platforms in general how do they address that?
And where does OpenSource win above buying everything in?
Community and open source can fix that
Ultimately as a vendor we need to come together with everyone and raise the bar for all
i.e. MikeT regularly converses with NetApp and other vendors to improve the overall quality of storage code in cloudstack
It means everyone can take advantage of these improvements regardless of
So going back to the things we defined earlier what particularly does storage need to offer to enable people to move to these highly automated, “As a Service platforms”? And does it really matter any more when you are buying things as a service?
You need the same flexibility and efficiencies from your hardware that you gain from your Cloud Automation platform
Otherwise you’re only part way there
Trying to force fit products that aren’t designed to work rarely ends well
I hear this debate frequently here and its why meetups like this are so useful
These 5 elements are what solidfire have defined as the benchmark and that we have built into our product.
I am going to focus on just the first three today.Personally I think these are the most critical – Scale – Guaranteed/Predictable performance and automated management.
I think they are what matter for storage that goes into CloudStack.
Whether you like what we do, or use someone elses storage, these are the things I would be push to bring into your platform.
Rack growth vs single system growht
SolidFire Scale OUT
Limited to the performance of the controller design
Controller upgrade, or new controller pair, to increase performance
New controller = data migration
Sprawling controllers = managing more islands of storage
Prone to noisy-neighbor problems & performance degradation
Controller based shared-disk systems can be SPF w/o full redundancy.
Generational upgrades are huge headache
Scale UP
Performance doesn’t degrade as capacity is added
Scale UP or DOWN by adding and removing nodes
No data migration, no downtime, no increase in management burden
Incrementally scale w/wide range of capacity and performance points
No disks shared between shelves, no SPF
Self-heal without the requirement for “extra” redundant components
Simple hardware upgrades, easily incorporate new hardware
To deliver predictable and guaranteed storage performance, SolidFire leverages a technology called Performance Virtualization which virtualizes performance resources in very much the same manner that capacity is virtualized today.
Patented by SolidFire, this technology allows administrators to manage storage performance separately and independently from storage capacity.
With SolidFire IT managers have unprecedented control over their storage system, and are able to deliver predictable storage performance to thousands of applications within a shared infrastructure.
Because performance and capacity are managed independently, you can create small volumes with extremely high performance, and very large volumes with very low performance. This allows you to provide predictable storage performance to a very broad number of applications.
On the right is a view of how volume level performance is set within the SolidFire storage system
Settings can be established for Min, Max, and Burst ensuring that applications get exactly the performance they require
(next slide)
Yet what makes SolidFire storage management really compelling is the ability to automate every aspect of the storage system.
At the core of the product is a complete REST-base API, which enables not only complete automation but deep integration of SolidFire capabilities such as the management of storage performance separate from storage capacity, and the ability to articulate SolidFire’s fine-grain storage performance controls
…and through this API we provide deep integration with the leading cloud orchestration and monitoring tools to make it easy for you to manage and provision storage resources, gather detailed usage data and granularly monitor and report on each component and user.
Its not just automation that matters though, you need simplicity in the first place.
I found that storage became quite complicated and while there very beneficial tools that vendors had developed around backup were useful, they often stifled platform innovate or create lock-ins. At SolidFire we focus on the simplicity and letting people get value from the elemental changes in our platform.
If you don’t have to design many of the things above it leaves you free to innovate elsewhere in your stack. It means you aren’t locked in.