Ignite 2015: Stretching failover clusters in WS2016

Session: BRK3487

Note: This session was jam packed with slides, text, and diagrams. The speaker was also flying through the material, so it was impossible to attempt to keep up. The session was very good, and quite technical. So if you deal with clustering in your daily job, check out the session recording for a boatload of good info.

  • Stretch clusters can achieve low RPO and RTO
  • Disaster avoidance is the new trend
  • Considerations when stretching clusters: Networking, storage

Recommendations: Adjust intra-node heartbeat thresholds; understand

Cloud Witness in Windows Server 2016

  • Leverages Azure as arbitration point
  • Quorum configuration achieved without an extra site
  • Writes a single blob per cluster
  • Costs on Azure is extremely low…in terms of pennies
  • Newly recommended quorum option

Storage Considerations

  • Storage replica is a brand new feature in WS2016
  • Block-level, volume-based synchronous & async using SMB 3.1.1
  • Any Windows volume, any fixed disk storage, any storage fabric
  • Baked into Windows..no need for third party storage

Hyper-V and General use file server are the main use cases for the tech preview. Not for SoFS.

Requirements & Recommendations

  • Datacenter edition & Azure stack SKUs only
  • Requires Active Directory (no schema updates, just Kerberos)
  • >1Gb network between servers
  • Disks: Must be GPT not MBR.
  • Free space on logs on NTFS/ReFS volume
  • Disk physical sector sizes must be the same (e.g. can’t mix 512e & 4K)
  • Network latency: 5ms round trip
  • Reality: 30-50Km apart
  • Network bandwidth is based on IO of the app and IOPS
  • Log volumes recommended on Flash (SSD, NVMe, etc.)
  • These are *strong* recommendations
  • Supports running inside a VM

Ignite 2015: Platform Vision: Storage

Session: BRK2485

What is SDS?

  • Cloud-inspired infrastructure and design – Commodity HW, cost efficiencies
  • Evolving technologies – Flash, network performance, VMs and containers
  • Data explosion – Device proliferation, modern apps
  • Scale out with simplicity – Integrated solutions, rapid time to solution, policy-based management

Storage Customer choice: Private cloud with traditional storage (SAN/NAS), private cloud with Microsoft SDS (Azure stack storage), hybrid cloud storage (StorSimple with Azure storage), public cloud storage (Azure storage) Where are we today in the SDS journey? -Goes into details about WS2012 R2 solutions, including those from Dell. What’s in Technical Preview 2 of WS2016?

  • Reliability – Cross-site availability and DR, improved tolerance to transient failures
  • Scalability – Manage noisy neighbors and demand surges, deploy mixed workloads in shared environment
  • Manageability – Easier migration to the new OS version
  • Reduced cost – More cost-effective by using volume HW, use SATA and NVMe in addition to SAS

Storage QoS

  • Simple out of box behavior – Enabled by default on scale out file server, automatic metrics per VHD, VM, host, volume; includes normalized IOPs and latency
  • Flexible and customizable policies – Policy per VM, VHD, service or tenant; fair distribution within policy
  • Management – System Center VMM and Ops Manager; PowerShell built-in for Hyper-V and SoFS

Rolling Upgrades

  • Simple – Rolling upgrade within cluster
  • Seamless – Zero downtime for Hyper-V and scale-out file server

VM Storage resiliency: reliability

  • Resiliency – Freezes VMs if storage path is lost
  • Visibility –

Storage Replica

  • Protection of key data and workloads
  • Synchronous replication – Storage agnostic mirroring with crash-consistent volumes
  • Increase resilience – Metro distance clustering
  • Complete solution – End-to-end for storage and clustering
  • Streamlined management – GUI management for nodes and clusters

Storage Spaces Direct

  • Cloud Design points and management – Standard servers with LOCAL storage, supports NVMe
  • Reliability, scalability, flexibility – Fault tolerance to disk, enclosure and node failures
  • Use cases – Hyper-V IaaS storage, hyperconverged
  • Partners: HP Apollo 2000, Quanta D51PH, Lenovo x3650, Dell, Cisco
  • Uses ReFS as the underlying filesystem

Azure-consistent storage (future preview)

  • Consistent – Azure consistent blob, table, and account management for on prem
  • Integrated – Deployed as Microsoft Azure stack cloud services
  • Manageable – Azure cmdlets, APIs and templates
  • Scalable – PaaS services as front-end, high-performance scale-out IaaS via internal SMB

Demo shows 100Gb NIC and NVMe Direct Storage with <1ms latency and 11GB (bytes) throughput over 1 network port. WOW! They literally saturated a 100Gb NIC with SMB traffic. Approximately 20% CPU utilization.