vmworld europe 2014: storage drs - deep dive and best practices
TRANSCRIPT
Disclaimer
• This presentation may contain product features that are currently under development.
• This overview of new technology represents no commitment from VMware to deliver these features in any generally available product.
• Features are subject to change, and must not be included in contracts, purchase orders, or sales agreements of any kind.
• Technical feasibility and market demand will affect final delivery.
• Pricing and packaging for any new technologies or features discussed or presented have not been determined.
CONFIDENTIAL 2
Storage DRS Deployment:
Thin Provisioning
Deduplication
Auto-Tiering
Outline
Storage IO Control Capabilities
New Control: IO Reservations
Storage DRS
Storage IO Control
New ESX IO Scheduler
New Control: IO Reservations
Storage DRS Integration
Array-Based ReplicationStorage DRS Integration
vSphere Replication
vSphere 6.0 Beta
• Features we are presenting here are available in beta
• Perfect time to get your voice heard
https://communities.vmware.com/community/vmtn/vsphere-beta
4
The Problem
What you see
Online store:
Product Catalog
Back-up
(low priority)
Shared
Datastore
Online Store:
Order Processing
What you want to see
Shared
Datastore
Online store:
Product Catalog
Back-up
(low priority)Online Store:
Order Processing
5
Storage Performance Controls
Shares: Relative importance of VMs
– IOPS will be allocated in this proportion
Limit: Maximum IOPS allowed per VM
Reservations: Minimum IOPS per VM
IOPS: 1200
600
600
IOPS: 1200
400
400
400
Online store:
Product Catalog
Back-up
(low priority)Online Store:
Order Processing
R: 100L: 500S: 20S: 100S: 100
Outline
New ESX IO Scheduler
New Control: IO Reservations
ESX 5.5 I/O Scheduler (mClock)
• Allocation of A and B will be in 1000:3000 ratio subject to reservation and limit constraint.
L: ∞
S: 3000
R: 100
L: 500
S: 1000
R: 200 Capacity (IOPS) VM A (IOPS) VM B (IOPS)
1000 250 750
500 200 300
200 133.33 66.66
A B
8
ESX 5.5 I/O Scheduler (mClock)
Compared to previous scheduler:
– Supports Reservation, Shares, and Limits
– Break large I/Os in 32KB
Limit: 100 IOPS
IO Size: 64KB
100 IOPS
Limit: 100 IOPS
IO Size: 64KB
50 IOPS
ESX 4.1 ESX 5.5
ESX 5.5 I/O Scheduler (mClock)
Outline
Storage IO Control Capabilities
New Control: IO Reservations
Storage IO Control
New ESX IO Scheduler
New Control: IO Reservations
Shared Storage
S: 100
SIOC
Local I/O Scheduler
Local I/O Scheduler+
Storage I/O Control
S: 100 S: 100 S: 100 S: 100 S: 100
1200 IOPS600 IOPS 600 IOPS 1200 IOPS
600300300 400 400 400300 300600
Storage IO Control
Control Congestion in shared datastore
Detect Congestion
– SIOC monitors average IO latency for a datastore
– Latency above a threshold indicates congestion
SIOC throttles IOs once congestion is detected
– Control IOs issued per host
– Based on VMs shares, reservations, and limits on each host
– Throttling adjusted dynamically based on workload
• Idleness
• Bursty behavior
Congestion Threshold
Datastore overload
Congestion threshold value (ms):
– Higher is better for overall throughput
– Lower is better for latency
Changing default threshold:
Percentage or absolute value
Default: 90% of peak IOPs capacity
Thro
ughput
(IO
PS
)
Datastore Load
No benefit
beyond certain
load
Late
ncy
Datastore Load
SIOC and Reservations
SIOC needs to be aware of VMs reservations
– The queue depth allocation of each host depends on the VM reservation
Backward compatible (ESX 5.5 or below)
Are reservations always satisfied?
– IOPS too low due to low latency threshold
– Background operations or errors in the array
In case reservations are not satisfied:
– SIOC will notify Storage DRS for further action
R: 500
SIOC
R: 500 R: 500
Outline
Storage IO Control Capabilities
New Control: IO Reservations
Storage DRS
Storage IO Control
Storage DRS:
IO Reservations
Functionality Overview
Storage DRS and IO Reservations
• Monitor SIOC reservation enforcement
– Migrate workload if reservations are not met
• Balance reserved IOPS usage in cluster
– Match datastore IOPS capacity to reservations
• Reservations as VM placement constraints
– Hard or soft constraints
• Per-datastore reservable IOPS
– Manual override allowed
Datastore Cluster
17
Ease of Storage Management
Initial Placement
Out of Space Avoidance
IO Load Balancing
Virtual Disk Affinity (Anti-Affinity)
Datastore Maintenance Mode
Add Datastore
Storage DRS Functionality: Brief Overview
Datastore
Cluster
Storage vMotion
•••
Storage DRS Deployment:
Thin Provisioning
Deduplication
Auto-Tiering
Outline
Storage DRS
Datastore Capacity Management
Used Space : 700 GB
Thin VMDK: 100 GBThick VMDK: 500 GB
Datastore Free: 1000 GBDatastore Free: 500 GBDatastore Free: 400 GB
Thin VMDK: 100 GB
Datastore Free: 300 GB
Storage DRS Configuration Option: PercentIdleMBInSpaceDemand, default = 25%
Provisioned Space : 1500 GB
Space Demand : 900 GB
20
Datastore: 2 TB
Thin Provisioned Datastores
Storage Array
Datastore: 1 TB
Bac
Datastore: 1 TB
Capacity : 4 TB
Used Capacity : 4 TB
Provisioned Capacity : 12 TB
Allocated Capacity : 6 TB
Virtual Disks
Backing Storage Pool
21
How does Storage DRS handle space over-commit?
• Hypervisor Level: Storage DRS manages space over-commit
– Fine grain control for overcommit magnitude
– Out of space avoidance using storage vmotion
• Datastore Level: Storage arrays manage space over-commit
– Storage DRS manages logical space usage in datastores
– VASA integration (v1): Storage DRS handles space outage signal from backing pool
– VASA integration (v2): Storage DRS controls space usage in backing pools
• Best practice: can use either method, but not both
22
Deployment with Deduplication
• Provides space efficiency
• Dedupe pool can span across multiple datastores
Dedupe
Storage DRS uses free space in datastore
VASA Integration (v2) : Storage DRS manages
logical space while keeping virtual disks in the
same dedupe pool
⤬Problem: Datastore appears to store
more data than capacity!
Total Virtual Disk
Size: 4TB
LUN Capacity: 1TB
23
Deployment with Auto-Tiered Arrays
• Multiple storage tiers
• VM data across tiers
• Tier use changes with workloadCapacity Tier
Performance Tier
Logical LUN of Auto-tier Array
Storage IO Control
IO priority
IO over-load remediation
Automatic initial placement
Space load balancing
Rule enforcement
Maintenance mode
Latency
IO Load
Auto-tiered array
SIOC Threshold
SDRS Threshold
24
Fine Grain Controls
All aspects of Storage DRS can be
controlled to suit your environments
Storage DRS Deployment:
Thin Provisioning
Deduplication
Auto-Tiering
Outline
Storage DRS
Storage DRS Integration
Array-Based Replication
vSphere SRM: Array-based Replication
• Storage DRS identifies replicated datastores
• All recommendations are in sync with replication policies:
– Automated moves within the same consistency group
– Manual moves for all VMs residing on replicated datastores
• Accounting of replication overhead due to Storage vMotion
27
Outline
Storage DRS
Storage DRS Integration
Array-Based ReplicationStorage DRS Integration
vSphere Replication
vSphere Replication (VR)
• Storage DRS discovers VR-replicas in datastores
• Storage DRS understands space usage of replica disks
• Storage coordinates moves with VR
– Space balancing
– Maintenance mode
29
Storage IO Control Best Practices
Avoid mixing vSphere LUNs and non-vSphere LUNs on the same physical storage
– SIOC will detect this and raise an alarm
Configure host IO queue size with highest allowed value
– Maximum flexibility for SIOC throttling
Keep congestion threshold conservatively high
– Will improve overall utilization
– Set lower if latency is more important than throughput
Datastore Cluster Best Practices
Similar datastore performance
May not be identical
Similar capabilities
Data management
Backup
✔Cluster1: Wide Perf
Variation
Cluster2: Similar
Datastores
31
Datastore and Host Connectivity
• Maximum possible host and datastore connectivity
• Improves DRS and Storage DRS performance
Partially Connected Datastore Cluster Fully Connected Datastore Cluster
More datastores in cluster better space and I/O balance
Larger datastore size better space balance
DRS Cluster DRS Cluster
32
vSphere Storage Policy based Management
Silver Disk Pool Gold Disk Pool
Data
store1
Data
store2
Data
store3
Data
store4
Cluster-1 (Tier2 VMs) Cluster-2 (Tier1 VMs)
Datastores with any storage
profileSilver Disk Pool Gold Disk Pool
Data
store1
Data
store2
Data
store3
Data
store4
Cluster-1 (Tier1 + Tier2 VMs)
Datastores with identical
storage profile
33
Summary
• Easier Storage Management
• Effective storage capacity and performance management
– Support for IO reservations
– Integration with storage environments
– Integration with vSphere solutions
• Many exciting features available in vSphere 6.0 Beta!
34
https://communities.vmware.com/community/vmtn/vsphere-beta