Pure Storage and VMware Integration Stefano Pirovano System Engineer @StePir75 © 2013 Pure Storage, Inc. | 1 Pure Storage is Flash for the Enterprise Consistent performance 100% MLC Flash Less cost than disk Inline deduplication & compression Mission-critical reliability 99.999%, non-disruptive operations Scalable & compatible 10 à 100s of TBs, Purity software © 2013 Pure Storage, Inc. | 2 DB, VSI, VDI: Where Flash & Dedup’ are disruptive Market-Leading All-Flash Array Gartner Magic Quadrant: Solid State Arrays, August 2014 #1 All-Flash Array for Databases, VSI, VDI Gartner Critical Capabilities: Solid State Arrays, August 2014 © 2013 Pure Storage, Inc. | 3 The Disruption of Simplicity: 5 Differentiators All-Inclusive Software Pricing Industry’s Broadest End-to-End Guarantee No Required Training or Professional Services CloudAssist! Real-Time Monitoring! Global Analysis / Analytics! Proactive Resolution! Continuous Improvement! Fanatically Proactive Support A Better Approach to Storage Acquisition & Lifecycles © 2013 Pure Storage, Inc. | 4 Agenda • VMware • Pure • Site VAAI Storage vSphere Web Client Plugin Recovery Manager • vVol Program © 2013 Pure Storage, Inc. | 5 vStorage API for Array Integration (VAAI) • SCSI-based offloading of common operations to storage array • Cloning of VMs, zeroing of disk space, metadata locking • Four • • • • supported primitives: Block Zero (WRITE SAME) Hardware-assisted Locking (ATS) Full Copy (XCOPY) Dead Space Reclamation (UNMAP) © 2013 Pure Storage, Inc. | 6 Hardware-Accelerated Zero • Without API – SCSI Write - Many identical small blocks of zeroes moved from host to array for MANY VMware IO operations – Pure automatically ignore zeros and never write them to the drives, so no “zero reclaim” penalty – New Guest IO to VMDK is “pre-zeroed” • With API – SCSI Write Same - One giant block of zeroes moved from host to array and repeatedly written – Thin provisioned array skips zero completely (pre “zero reclaim”) SCSI WRITE (0000…..) SCSI WRITE (data) SCSI WRITE SAME (0 * times) SCSI WRITE (0000….) SCSI WRITE (data) SCSI WRITE (data) Repeat MANY times… VMFS-5 VMDK • Use Cases – Reduced IO when writing to new blocks in the VMDK for any VM – Time to create VMs (particularly FT-enabled VMs) © 2013 Pure Storage, Inc. | 7 Hardware-Accelerated Locking • • • Without API • Reserves the complete LUN so that it could obtain a lock • Required several SCSI commands • LUN level locks affect adjacent hosts With API • Locks occur at a block level One efficient SCSI command - SCSI Compare and Swap (CAS) • Block level locks have no effect on adjacent hosts. VMFS-5 Use Cases • Bigger clusters with more VMs • View, Lab Manager, VMware vCD • More & Faster VM Snapshotting © 2013 Pure Storage, Inc. | 8 Hardware-Accelerated Copy • Without API – – – – SCSI Read (Data moved from array to host) SCSI Write (Data moved from host to array) Repeat Huge periods of large VMFS level IO, done via millions of small block operations “let’s Storage VMotion” SCSI READ SCSI READ SCSI EXTENDED SCSI READ COPY ..MANY times… • With API – – – – SCSI Extended Copy (Data moved within array) Repeat Order of magnitude reduction in IO operations Order of magnitude reduction in array IOps VMFS-5 SCSI WRITE SCSI WRITE SCSI WRITE ..MANY times… VMFS-5 • Use Cases – Storage vMotion – VM Creation from Template “Give me a VM clone/deploy from template” © 2013 Pure Storage, Inc. | 9 FlashRecover Accelerated VM Cloning Hardware-Driven VM Cloning via XCOPY Legacy Disk Array (no XCOPY) Legacy Disk Array (with XCOPY) Pure Storage Accelerated VM Cloning Virtualization Virtualization copy Virtualization Metadata Snap copy Stored Twice Stored Twice Stored Once 112 Seconds 22 Seconds 10 seconds Example: Comparing Pure Storage vs. xxxxxx cloning 1,000 40GB VMs* * http://virtualgeek.typepad.com/virtual_geek/2012/12/vmax-and-vsphere-vaai-xcopy-update.html © 2013 Pure Storage, Inc. | 10 Hardware-Accelerated Copy 100 GB Zeroedthick Virtual Disk (50 GB of data) © 2013 Pure Storage, Inc. | 11 Hardware-Accelerated Copy © 2013 Pure Storage, Inc. | 12 Hardware-Accelerated BlockDelete • • • Without API • Changes to vmdk is not propagated to backend LUN (how would I know!) • If data is constantly rewritten, LUN would grow till it fills up With API • Deleting blocks gets translated to block deletes on LUN • Better LUN space management Use Cases • Great for us as we can mark the deleted blocks and delete immediately – Less confused VI admin • No backend LUN bloat © 2013 Pure Storage, Inc. | 13 Flash Makes Virtual Administration Faster 3 min 15 min Disk Disk 10 sec <1 min Provision 50GB VM From Template (XCOPYBenefits) Boot 100 VDI desktops (ATS benefits) 4 min 45 min 4 min SvMotion 50GB VM (XCOPYBenefits) Disk Disk 10 sec Recompose 100 VMs © 2013 Pure Storage, Inc. | 14 Complete Management within vSphere vCenter Web Client Plugin Complete management of FlashArray from within vCenter. Automated Datastore Creation Just specify size and done. No LUNs, no RAID, no WWNs, no rescanning. Complete Capacity Visibility See through deduplication, compression, and thin provisioning to understand real capacity. Complete Performance Visibility Correlate IOPS, latency, and bandwidth on a per-datastore basis. VMware Ready and VAAI Certified Highest performance possible, jointly supported by Pure Storage and VMware. © 2013 Pure Storage, Inc. | 15 Pure Storage vSphere Web Client Plugin © 2013 Pure Storage, Inc. | 16 Datastore à Array Visibility © 2013 Pure Storage, Inc. | 17 Automated Datastore Creation © 2013 Pure Storage, Inc. | 18 Automated Datastore Resizing © 2013 Pure Storage, Inc. | 19 VMware Site Recovery Manager • Disaster Recovery automation product for VMware environments • Leverages array-based replication to migrate or recovery virtual machines from one datacenter to another • Interacts with vendor-supplied Storage Replication Adapters (SRAs) to migrate storage • No licensing costs for Pure Storage replication or SRA use © 2013 Pure Storage, Inc. | 20 FlashRecover Replication The Benefits of Asynchronous + Snapshot Replication Combined Flexible replication with low RPO … 1000s • Differentials-based, bi-directional • Replicates every minute (configurable) • Retains library of PIT remote snaps for recovery Simple setup in minutes • Automation with Protection Policies (consistency groups and variable retention) • No pre-planning or professional services! Data reduction-optimized • Always thin, deduped, compressed • Delta changes only after baseline • Data reduction accelerated Clone replicas Instant recovery for Zero RTO • Instantly export any PIT replica • Instantly roll backward or forward Advanced Multi-Site Replication • 1:Many, Many:1 or Many:Many © 2013 Pure Storage, Inc. | 21 FlashRecover Protection Policies Policy-based Automation of Local and Remote Retention Schedules Target Source Day 1 Day 0 Volume Volume(s) Host(s) Host Group(s) Protection Groups • Delivers data protection consistency across multiple volumes or multiple hosts • Group multiple objects within a Protection Group • Objects can belong to multiple Protection Groups Retain all replicas Day 1 Day 0 Day 2… Retain all replicas Variable Retention Automation • Configurable frequency schedule and retention period for local snapshots and replication • Policy automatically creates and expires snapshots and remote replicas © 2013 Pure Storage, Inc. | 22 Recovery with FlashArray Protected Environment Recovery Environment Snapshot A1 Remote Snapshot A1 Snapshot A2 Remote Snapshot A2 Volume A Snapshot A3 Volume B Remote Snapshot A3 © 2013 Pure Storage, Inc. | 23 vVols and VASA • What are vVols and VASA? • What problems are they solving? • What opportunities do they create for Pure? • What risks do they present? • What are we planning to deliver? © 2013 Pure Storage, Inc. | 24 Old world • Create a volume • Export to a bunch of ESXi hosts • Create a VMFS filesystem and attach it to vCenter as a datastore • Create VMs or migrate them onto the datastore © 2013 Pure Storage, Inc. | 25 Problems • VMFS is slower than raw device • Duplication of functionality – both storage array and ESXi implement thin provisioning, fast cloning, and snapshots • • • Storage array does it on a volume level ESXi does it on a virtual disk level Storage array functions apply to all VMs in the LUN – often want to target a single VM • Have a big pool of storage that you want to share among many ESXi servers • • Have to carve it up into multiple LUNs because of clustering limits in VMFS Get the division wrong, it can be hard to fix § Can’t shrink a VMFS volume § Expanding may fail under load © 2013 Pure Storage, Inc. | 26 Solution - vVols • Give • each virtual disk of a VM its own volume E.g. 2000 VMs with 2 virtual disks each = 4000 volumes • Expose • • • • • a standard API (through VASA) for Create/resize/delete a volume Snapshot volume Clone volume List volumes Connect/disconnect volume to host © 2013 Pure Storage, Inc. | 27 Important details • What • • about small config files (e.g. .vmx)? They live in a per-VM config volume (2-8GB) Happens to be formatted VMFS • Each snapshot or clone of a vvol is a separate vvol © 2013 Pure Storage, Inc. | 28 Old vs. new world VMFS volume FlashArray FlashArray © 2013 Pure Storage, Inc. | 29 vVols - Problems solved • Snapshot/clone of virtual disk ó snapshot/clone of volume • No VMFS translation overhead for a VM’s virtual disks • No shared VMFS metadata that needs co-ordinated updates == fewer clustering limits • No VMFS filesystem size to limit the size of the VMs (double-edged sword) • Fewer XCOPYs / UNMAPs – instead we see snapshot/ delete volume © 2013 Pure Storage, Inc. | 30 vVols - opportunities • Per • • virtual-disk or VM services Space and perf reporting Replication / data protection / tiering • Simpler user experience © 2013 Pure Storage, Inc. | 31 Storage containers • Motivation • • Many storage arrays have different pools (RAID-6 pool vs RAID-10 pool) Want to target vvol creation to the right pool • Solution • • • • Each pool can expose a storage container to VMware Storage containers are mapped to datastores When creating/cloning a virtual disk, admin can choose a target datastore ESXi targets the create/clone operation to corresponding storage container © 2013 Pure Storage, Inc. | 32 Storage containers & Pure • We’ve got one pool so we only need to expose one storage container • But: • • • • Could allow users via our GUI to create additional storage containers for organization (finance vs. engineering) When replication vvols from a remote array, might want to put them in a separate storage container for easy clean-up in case replication arrangement ends Might need a storage container per vCenter - unclear whether 2 vCenters can safely share a storage container Might be useful when we have quotas © 2013 Pure Storage, Inc. | 33 About those new APIs • Create/resize/snapshot, • Didn’t etc. extend SCSI or NFS to add these features • Added the APIs to VMware APIs for Storage Awareness (VASA) • VASA is accessed over HTTPS/TCP/IP/Ethernet through the management port • If management port is down, you can’t create or start VMs! • Today, if IP is down, a FibreChannel array will still happily serve data. © 2013 Pure Storage, Inc. | 34 VASA – grab bag of features • vVol APIs • Describe the storage array to vCenter • • • How many controllers? How many ports on each controller? What logical units are exposed? What is the service level of logical unit (e.g. gold, bronze)? Supported protocols? Like an SNMP or CIM object model Tell VMware about what storage pools are backing the LUNs it sees • Don’t automatically do storage vmotion across these two LUNs because they are in the same pool • Policy APIs • Alert / event API for raising alerts/events with vSphere © 2013 Pure Storage, Inc. | 35 Policies • Problem • VMware administrator is creating a VM – which datastore should he use? Today: sends e-mail to storage admin – what are the SLAs associated with the various datastores? • How does VMware administrator find out if the SLA isn’t being met/has changed? • • Solution – VASA policies • • • • Schema – vendor can programmatically describe the knobs supported (e.g there’s a replication Interval knob and it’s a time duration) Profile on vVol – requested knob settings (e.g. this vVol needs a replication Interval of 5 minutes) Profile on storage container – knobs supported by storage container (replication Interval 5 min to many minutes) Compliance – vCenter periodically queries the storage and asks whether it has been able to satisfy the policies © 2013 Pure Storage, Inc. | 36 Policies – new world • Admin • • • • creates virtual disk Specifies the policy they want to apply VMware shows compatible datastores Admin chooses one Admin can get alert if datastore is unable to meet policy © 2013 Pure Storage, Inc. | 37 Thanks!