Nov 19, 2014
VMware + SolidFireDeliver Predictable Application Performance & Improve Infrastructure Efficiency
Aaron Delp, Cloud Solutions Architect
▪ Intro to SolidFire (Short Version)
▪ Consolidation of Multiple Mission Critical Applications
▪ Integrated End-to-End Quality of Service (QoS)
▪ End-to-End QoS Demo
▪ Eliminate VM Sprawl & Wasted Resources
Agenda
Intro to SolidFire(Short Version)
The SolidFire SolutionScale-out high performance storage systems designed for large scale infrastructure
Most Scalable All-Flash storage system 5 – 100 nodes, 60TB – 3.4PB, 7.5M IOPS Industry-standard hardware, 10 GigE iSCSI 20X performance of traditional SANs
Most complete enterprise feature set of any AFA
10x reduction in operational cost
Delivered below $3/GB
Deploy new applications and capabilities faster
Provide more agile and scalable infrastructure
Increase application performance and predictability
Enable automation and end-user self-service
Raise operational efficiency and reduce cost
Enterprise IT lacks storage agility, and is
under significant pressure
Deploy new applications and capabilities faster
Provide more agile and scalable infrastructure
Increase application performance and predictability
Enable automation and end-user self-service
Raise operational efficiency and reduce cost
Otherall-flash arrays
only solve for performance
Deploy new applications and capabilities faster
Provide more agile and scalable infrastructure
Increase application performance and predictability
Enable automation and end-user self-service
Raise operational efficiency and reduce cost
SolidFire solves for the
Next Generation Data Center
Scale-Out Architecture
New nodes are added as demand dictates Performance and capacity instantly available to all volumes Nodes added on the fly without down time
174 TB
207 TB
241 TB
275 TB
375,000 IOPS
450,000 IOPS
525,000 IOPS
600,000 IOPS
Performance
Cap
acity
Linear ScalabilityStarter 5 node SF9010 configuration
5, SF9010 Nodes 174 TB 375,000 IOPS
Scale-Out Architecture
Scale beyond 100TB
In-line Data Reduction
Shared Nothing HA
Guaranteed Performance
Automated Management
All Flash Storage
All Flash Arrays Are Not Created Equal…
Workload ConsolidationMultiple Mission Critical Applications on the Same Storage System
One of the major problems we are solving for…
Noisy Neighbors!
A few resource (IO) hungry applications negatively affect all other’s performance
Traditional Multi-Tenant Performance
Noisy Neighbor
Eliminating Noisy Neighbors with QoS
The Noisy Neighbor Effect Individual tenant impacts other applications
Unsuitable for performance sensitive apps
SolidFire QoS in Practice Create fine-grained tiers of performance
Application performance is isolated
Performance SLAs enforced
Noisy Neighbor
Performance 0
Performance 1
Performance 2
Performance 3
Decreased Performance
Performance Virtualization: Unified 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
Guaranteed Performance - QoS
Enabling Delivery of all Applications from a Single Storage Infrastructure
Scale-Out ArchitectureGuaranteed PerformanceFully automatedBelow $3/GB
VDI
Integrated End-to-End QoSVMware SIOC + SolidFire QoS = Predictable Performance
Hypervisor-based QoS? (SIOC)
Tries to achieve “fairness” between VMsAddresses some of the performance variability of VMs
But assumes that storage is consistent
never fails
is not used by anything else
An environment that depends on predictable performance demands a more coordinated approach across host & storage resources
Lack of control A hypervisor can Throttle IO Prioritize IO in its own queue
But a hypervisor cannot Control or maintaining the total IO pool available or make
more IO available Force the storage system to deliver a specific level of
performance
Any OS or hypervisor will always be at the mercy of the storage device abilities and state
Performance Degradation
The hypervisor does not know the cause of performance degradation. Is it: A noisy neighbor? A failure condition? A maintenance in process? Resources removed or added to the storage?
SIOC’s response is to turn up the volume (queue) on high priority apps & drown out the noisy neighbor Could make the problem much worse Hopefully the degradation cause is not actually a higher
priority Hopefully two independent SIOC decisions are never made
As storage system utilization increases performance degradation becomes a larger concern
Inability to set and guarantee minimum level of performance
It’s easy to cap or prioritize performance
But a lot can get lost in translation?
SIOC Priority? Shares? Apps and tenants need a
minimum level of performance articulated in a meaningful way
IOPS, BW, and Latency
▪ Consolidate all use cases (e.g. cloud infrastructure, VDI, virtualized apps) with confidence▪ SIOC with storage enforced QoS ensures predictable performance to each VM▪ Automated, dynamic performance allocation to every datastore
End-to-End QoS
VM QoSDatastore QoS (Tiering)Volume QoS
QoS managed per
SolidFire Volume Non-SIOC enabled
environments
QoS managed per
vSphere datastore Configure SIOC
defaults per datastore Manage hundred’s to
thousand’s of VM’s
without setting SIOC
per VM Storage vMotion to
change performance
QoS managed per VM Configure SIOC per
VM Granular flexibility
regardless of
datastore location to
manage individual
VM’s Edit VM settings to
change performance
Flexible & Powerful Performance Control Options
End-to-End QoS Demo
Eliminate VM Sprawl & Wasted Resources
Forced Over-Provisioning
Trying to stay away from degradation many are forced to massively over-provision
Only way on most systems to ensure a large enough IOPS pool with no contention
Underutilization kills economics of shared storage environment.
What is your usable $/GB factoring in underutilization? Not uncommon to find SANs dedicated to an app
Always On, In-Line Efficiency
All data is compressed, de-duplicated and thin provisioned
Accomplished without any performance impact
Part of system architecture, not bolt-on feature
Operational Expense reduction in addition to Capital Expense reduction
30TB 60TB 90TB 120TB Provisioned
Purchase
15TBRaw
15TBHelix DP
Thin Provisioning
Effective Capacity
15TBHelix DP
Compression
Effective Capacity15TBHelix DP
De-duplication
Effective Capacity15TBHelix DP
Snapshots & Clones
Efficient versioning Deduplicated and compressed in-line Maximizing capacity Minimizing network traffic
Enhancing durability and availability Any snapshot can be cloned As a separate mountable, deduplicated volume As a real-time backup to another any Swift or S3
API compatible target
Operational Efficiency (compares 173TB of effective capacity)
Reduce rack space Consume less power And increase performance
52,000 IOPS
375,000 IOPS
Questions?
▪ Intro to SolidFire (Short Version)
▪ Consolidation of Multiple Mission Critical Applications
▪ Integrated End-to-End Quality of Service
▪ End-to-End QoS Demo
▪ Eliminate VM Sprawl & Wasted Resources
More Information: solidfire.com/vmware