In this webinar we will learn what the High Availability & Storage team in Microsoft has cooked up for us in Windows Server 2016, which is being launched at Microsoft Ignite at the end of September.
There’s lots of new stuff in this release, including better high availability for Hyper-V, greater control over resource utilization, improved fault tolerance of transient events, newer design options for stretch or multi-site clusters, a whole new way of doing software defined storage with SATA and NVMe drives, built-in block-level storage replication, and hyper-convergence without having to break the bank.
Whats new in Microsoft Windows Server 2016 Clustering and Storage
1. www.mwh.ie
What’s New in Windows Server 2016
Failover Clustering & Storage
Aidan Finn, MVP
Technical Sales Lead, MicroWarehouse
2.
3. www.mwh.ie
What’s New in Windows Server
2016 Hyper-V
Aidan Finn, MVP
Technical Sales Lead, MicroWarehouse
4. www.mwh.ie I
About Aidan Finn
• MVP, Cloud & Datacenter
Management (Hyper-V)
• Experienced with Azure, Hyper-V,
Windows Server/Desktop, System
Center, and IT infrastructure
• http://www.aidanfinn.com
• http://www.petri.com/author/aidan-finn
• @joe_elway
• aidanfinn.com
Technical Sales Lead, MicroWarehouse
5. www.mwh.ie I
About MicroWarehouse
• Irish owned/located distributor
• Park West, Dublin 12, Ireland
• Distributors for:
• Microsoft on-premises & cloud
• Microsoft Surface
• DataOn for Storage Spaces
• Gridstore for Hyper-Convergence
• SkyKick for Office 365 backup
• And many more
• Value added distribution:
• Much more than selling licenses
• Get your licensing right
• Sales education
• Technical training
@MWHDistribution
Value Added Distribution
7. www.mwh.ie I
Agenda
• Failover Clustering
• Storage Replica
• Storage Spaces Direct (S2D)
• Troubleshooting
• Some event news
What we are covering today
9. www.mwh.ie I
What Are the Talking Points?
• Cloud
• Inspired by Azure
• Easier management
• Improved service availability
• Feedback has shaped WS2016
• https://windowsserver.uservoice.com
• WS2012 and WS2012 R2
• 5 x WS2016 Technical Preview releases
Cloud, cloud, cloud
11. www.mwh.ie I
VM Start Ordering
• Configure sets of virtual machines
• Model dependencies of applications
1. Domain controllers
2. Database servers
3. Application servers
4. Web servers
• Create dependencies
• VM started
• Delay/wait
• Integration component heartbeat
• Define success of a set
• Example, 2 of 5 machines are running
Hyper-V
12. www.mwh.ie I
Storage Resiliency
• Most infrastructure problems are
transient
• Glitch on a SAN network that lasts seconds
• Delay in VM IO caused VMs to bug check
• Takes several minutes to resume services
• WS2016 adds Storage Resiliency
• VM placed in Paused-Critical state when IO stops
• Retains application state
• VM starts again when storage resumes
• Reduces impact on business
Hyper-V
13. www.mwh.ie I
Compute Resiliency
• Most infrastructure problems are
transient
• Glitch on a cluster network that lasts seconds
• 10 second glitch causes VM failover in a cluster
• Takes several minutes for services to resume
• WS2016 adds Compute Resiliency
• VMs aren’t moved if they are still on the network
• Repeat host issues cause the host to be
quarantined for 2 hours
• VMs are live migrated to healthy hosts
Hyper-V
14. www.mwh.ie I
Storage Resiliency
• Moving VM placement load balancing
from System Center to the core product
• Runs:
• Node joins the cluster
• Every 30 minutes
• Based on CPU and RAM
• 3 levels of aggressiveness:
• Low (>80%)
• Medium (>70%)
• High (Any host is 5% more than
average)
Hyper-V
15. www.mwh.ie I
Cloud Witness
• Multi-site cluster required a file-share
witness
• Preferable another cluster
• In a 3rd site
• Sites A & B each required connection to Site C
• Azure is a 3rd site
• Cloud witness
• Blob API on storage account
• Just a sequence number – no data
• In testing, no one has incurred enough charges for
billing!
Stretch Cluster
16. www.mwh.ie I
Site Awareness
• Make clustering a DR solution
• Provide site awareness to hosts
• Normal placement stays inside the site:
• Failover
• Load balancing
• VMs follow (Live Migration) CSV
ownership
• Prevents latent redirection going across the WAN
Stretch Cluster
17. www.mwh.ie I
Mixed Domain or No Domain
• A cluster has always required AD
membership
• WS2016 supports:
• All nodes in Workgroup
• Nodes being in different domains
• Great for SQL clusters
• Can be used by Hyper-V
• You lose Live Migration
Stretch Cluster
18. www.mwh.ie I
Storage Networking
• SMB Multichannel
• LBFO for SMB 3.x
• 20 Gbps instead of 10 Gbps
• Automatic failover
• Failover clustering required NICs to be
on different subnets/VLANs
• WS2016 allows cluster NICs to be on
the same subnet/VLAN
• Simplifies configuration
• Reduces VLAN provisioning
Easier design and greater scale for SMB 3.0
22. www.mwh.ie I
Challenge of Disaster Recovery
• Microsoft witnessed lots of enterprises
struggling with DR
• Had a SAN in Site A and in Site B
• Either had:
• No storage replication (license too expensive)
• Unsatisfactory replication
• Microsoft decided to implement storage
replication at the software layer
SAN replication licensing
23. www.mwh.ie I
Storage Replica Overview
• Implemented in software low in the
storage stack
• Replication not affected by
• Dedupe / thin provisioning
• BitLocker
• NTFS / ReFS
• SMB
• VSS
• Doesn’t care about the physical storage
• Two identical SANs
• Two different SANs (including manufacturers)
• Between Storage Spaces
• Storage Spaces to SAN (and vice versa)
Disaster recovery
24. www.mwh.ie I
Replication
• Block level
• Replicates volumes
• Only provisioned blocks (effective thin
provisioning)
• Synchronous replication
• Zero data loss
• Impact on storage latency
• Best with <5 ms links
• Up to 80 KM between sites
• See Mellanox Infiniband and MetroX)
• Asynchronous replication
• Long distances between sites (latency)
• Minimal impact on storage latency
What is happening
25. www.mwh.ie I
Seeding
• You can seed data
• How do you copy volumes to disk?
• SR includes an interesting capability:
• Replicate local volume A to local volume B
• Faster than RoboCopy
• RoboCopy cares about files – metadata
• SR is lower in the stack – files/metadata are
irrelevant
The first synchronization
26. www.mwh.ie I
Replication Networking
• Uses SMB 3.11
• Easier firewall management
• Can leverage SMB Direct (RDMA)
• iWARP
• ROCEv2 (routable ROCE)
• Infiniband (MetroX)
• You can define which network is used
for replication
• You can control this on a per replicated volume
basis
What is happening
28. www.mwh.ie I
Leveraging Servers & Commodity Disks
• Normal rack
servers
• Internal disks
• SAS / SATA
• HDD
• SSD (50,000 IOPS)
• NVMe (420,000+
IOPS)
• Required for
production:
• 10 Gbps networking
• RDMA / SMB Direct
Storage Spaces Direct
29. www.mwh.ie I
Cluster The Nodes
• Aggregate the
servers
• Up to 16 in a single
cluster
• In TP5
• Min of 3 (see later)
• High availability
• Leverage CSVFS
Storage Spaces Direct
30. www.mwh.ie I
Pool the Disks
• Aggregate the
disks
• A Storage Spaces
storage pool
• Capacity
• Performance
(IOPS)
• Fault tolerance
Storage Spaces Direct
31. www.mwh.ie I
Enable Storage Spaces Direct
• Enable-
ClusterStorageSpa
cesDirect
• Simplified setup
• Analyses disks
• Configures caching,
tiering, and capacity
Storage Spaces Direct
32. www.mwh.ie I
Enable Storage Spaces Direct
• Create virtual disks
(LUNS)
• Define tier sizes
• Configure disk fault
tolerance
• Format with ReFS
• Not NTFS
• Add as cluster
CSVs
• Start using the
cluster
Storage Spaces Direct
33. www.mwh.ie I
How We Can Use S2D Clusters
• Scale-Out File Server
• SAN alternative
• Hyper-V nodes store VMs on S2D cluster using
SMB 3.0
• New way to do SOFS instead of using JBOD
• Removes the SAS layer
• Adds possibility of cheaper SATA HDD/SSD
• You actually can have 1 JBOD per node for more
capacity
• Hyper-converged infrastructure (HCI)
• Each node is compute + storage
• No SMB 3.0 – just store VMs on CSVs
• Lego-block infrastructure
Storage Spaces Direct
34. www.mwh.ie I
Resiliency
• Disk fault tolerance
• Defined at virtual disk
(LUN) level
• Server fault tolerance
• Up to 2 nodes
• Rack fault tolerance
• Define fault domains
• Spread S2D nodes across
racks
• Disaster recovery
• Hyper-V Replica
• Azure Site Recovery
• Storage Replica
Storage Spaces Direct
35. www.mwh.ie I
Caching
• S2D supports SATA/SAS
HDD, SSD and NVMe
• Even a mixture of all 3
• Fastest disks in each
node
• Caching devices
• Used as read/write cache in that
node
• Remaining disks are
capacity devices
• Each capacity device bound to
1 caching device
• Used to create virtual disks
(LUNs)
Storage Spaces Direct
NVMe
SSD SSD SSD
HDD HDD HDD
NVMe
SSD SSD SSD
HDD HDD HDD
S2D Node
36. www.mwh.ie I
Multi-Resilient Tiering
• Virtual disk can feature
mixed disks (SSD and
HDD)
• Virtual disk can feature
mirroring AND parity
• 3-way mirror for 4+ nodes
• 2-way mirror for 3 nodes
• Writes sent to fast mirror
tier
• Cooler data sent, on the
fly, to slower/economic
parity tier
Storage Spaces Direct
Cache
3-Way Mirror Parity / Erasure Coding
Write
Demote aged blocks
ReFS update of modified blocks
38. www.mwh.ie I38
Storage Spaces hardware, distributed to European Union resellers
SMB or Branch Office
2-Node Hyper-V Cluster
12 x clustered data drives (e.g. 4 x SSD + 8
x HDD, up to ~ 32 TB usable tiered Hyper-
V storage)
1023W (1+1) redundant power
2 x clustered Hyper-V hosts, each with:
o 2 x Intel® Xeon® E5-2600v3 (Haswell-
EP)
o DDR4 Reg. ECC memory up to 512GB
o 2 x 1G SFP+ & IPMI management
“KVM over IP” port
o 2 x PCI-e 3.0 x8 expansion slots
o 1 x 12Gb/s SAS x4 HD expansion port
o 2 x 2.5” 6Gb/s SATA OS drive bays
SAN-Alternative for Hyper-V
Scale-Out File Server JBOD
24 x 2.5”, 32 * 3.5” or 70 * 3.5” JBODs
Up to 660 TB usable storage for Hyper-V
o 4 x 70 disk JBOD
o 10 SSD + 60 HDD per JBOD
o 3-way mirroring
o Enclosure awareness / fault
tolerance
• Present to Hyper-V cluster using Windows
Servers
• Super-fast SMB 3.0 connectivity
Coming Soon!
Storage Spaces Direct
• Hyper-Convergence using Windows Server
2016
• 1U and 2U models
• Leveraging NVMe and SSD flash storage
• A recent demo hit 2.4 million IOPS
• To be launched at Microsoft Ignite (end of
September)
40. www.mwh.ie I
Easily Get Diagnostic Data
• Data is in many logs across many
servers
• Failover clustering
• Storage Spaces
• Replication
• SMB / Networking
• Get-StorageDiagnosticInfo
• Start logging from one place
• Aggregate logs into 2 chronologically ordered files
• Event log + ETL
Troubleshooting
42. www.mwh.ie I
Value-Added Distribution
• Keep an eye on http://learn.mwh.ie
• Follow @MWHDistribution
• Or take our mailshots off your junk mail filter
We will do our best to help
43. www.mwh.ie I
Windows Server 2016 Launch
• October 26th – The Irish
Windows Server 2016 launch
• 10am in Dublin City
• Follow @MWHDistribution for
details
• Keynote + 2 breakout tracks
(on-prem + cloud)
• MVP speakers from around
Europe
• Including Didier Van Hoye,
Damian Flynn, Carsten Rachfahl,
Kevin Greene
Community launch by MicroWarehouse & MVPs
44. www.mwh.ie I
Azure IaaS VMs Hands-On Training
• Novmber 22-24th in Dublin
• 3 days of hands on, in Azure, IaaS VM
training
• Lots of practical work
• From the basics to JSON, all in ARM/CSP
• Storage, networking, peering/VPN, DMZ, VMs,
PowerShell, backup, load balancing, geo-fault
tolerance, and more!
• Contact events@mwh.ie for registration
& (low) costs details
3-day technical ARM/CSP training course
45. www.mwh.ie I
Thank You!
Aidan Finn
aidanfinn@mwh.ie
@joe_elway
aidanfinn.com
http://learn.mwh.ie
@MWHDistribution
Watch out for emails about future events!