Top Storage Considerations to Plan for Before Deploying VDI

DAN BOGGS, SENIOR PRODUCT MANAGER NEXENTA SYSTEMS Virtual Desktop Infrastructure

Key Metrics • End user – Latency – User experience highly depends on response time

• Service provider – $/user The VDI I/O Challenge

• Workload is 95% small random writes • Toughest workload for a storage system VDI Workloads

• Boot and login storms • Stateless sessions – Users login to any VM • Stateful sessions – Users return to the same VM

Recommendations for VM Config

• RAM per user to avoid heavy paging – 800MB – 1GB • Plan for more resources for Windows 7 clients – Significantly higher RAM requirements – 100 IOPs on average during boot Can’t assume customers will use – 40% more IOPs than XP performance optimized builds • Give VM two separate disks – One for paging – One for data Storage Considerations

• Storage system have support for…? – Allocate-on-flush architecture – Compression – Hybrid pools – Variable sizes – VSA support – De-duplication – Virus scanning Zettabyte

• 256-bit checksums Enables exceptional • Hybrid storage pools price/performance • Instantaneous snapshots • Thin provisioning Efficient use of storage • In-line compression • In-line and in-flight de-duplication

Originally developed by and part of OpenSolaris Variant of ZFS is now part of the open source .org ZFS Components

Discover logical disks Create datasets Determine RAID requirements Assign properties (e.g. de-dup) Create storage pools Share (NFS, CIFS, iSCSI, FC)

Pool-A Pool-B

NFS iSCSI CIFS NFS Hybrid Storage Pools Enables outstanding price/performance

Read Cache

Transaction log ZFS-based VDI Reference Architecture

• Configuration settings – Mirroring – Compression – Tunable for lower latency (vs. throughput) – Honor sync requests – 4k record sizes – Virus scanning here instead of VM Virtual Storage Appliance VSA approach offers simplification “VDI-in-a-Box”

Virtual Machines

ZFS as VSA

ZFS as VSA

Local Cache Hypervisor Backend Storage VSA Approach

• Package the storage as a VM on the hypervisor system – As you add hypervisors to scale, you add the storage VM as well – Uniform IOPs per user as you scale • Deliver I/O to the hypervisor from inside the hypervisor – Minimize network traffic • Reduces number of slots needed on the server Storage for Paging Files

• Load server with extra RAM • Create pool of mirrored SSDs • No separate ZIL needed • No separate read cache needed • Compression • No de-duplication Backend Storage

Storage for applications and home directories • RAIDz2 configuration • Mirrored, separate ZIL devices • Read cache – Configurable amount based on specific workloads • NFS recommended – Application-specific timeouts and recovery behaviors make iSCSI challenging – Enables scale-out via NFS referrals – Blocks can have locking contention with VMFS Recommendations for Stateless Sessions

• VDI clients are disposable – Can login to a different VM each time • Cache VMDK on solid-state disk • Can disable the ZIL to accelerate performance • Profile and personalization applications stored on central storage Recommendations for Stateful Sessions

• Put boot image in the SAN • With VSA, can mirror data disk with copy on VSA and copy in the SAN – ZFS will just use first response for reads

Summary

• Storage systems based on the ZFS architecture are well suited to address challenging VDI workloads Thank You!

DAN BOGGS, SENIOR PRODUCT MANAGER NEXENTA SYSTEMS