Nutanix Data Protection and Disaster Recovery Tech Note_final.pdf

©2014 All Rights Reserved, Nutanix Corporation

Nutanix Tech Note Data Protection and Disaster Recovery

Nutanix Virtual Computing Platform is engineered from the ground-up to provide enterprise-grade availability for critical virtual machines and data. This Tech Note covers the data protection and disaster recovery functionality of the Virtual Computing Platform.


Nutanix Virtual Computing Platform

This Tech Note discusses the data protection and disaster recovery functionality in the Nutanix Virtual Computing Platform. We also recommend reading the Nutanix Tech Note on system reliability to learn more on about the resiliency features of the Virtual Computing Platform including how hardware and disk failures are handled.

The Nutanix distributed software architecture runs a virtual storage controller (Controller VM or CVM) on each Nutanix node or host on the Virtual Computing Platform, forming a distributed system. All nodes actively work together to aggregate storage resources into a single global pool that can be leveraged by all. The storage resources are managed by the Nutanix Distributed File System (NDFS) to ensure that data and system integrity is preserved in the event of node, disk or application or hypervisor software failure. NDFS also delivers data protection and high availability functionality that keeps critical data and VMs protected and applications running.

Starting with Snapshots

The foundation of the Nutanix data protection and disaster recovery functionality is the concept of a VM-centric snapshot. To understand the advantage of Nutanix snapshot functionality, it is important to understand the different types of snapshots available today.

A snapshot is an evolution of the traditional backup process. It is created when the storage system creates a full or virtual copy of the metadata or the index of the stored data. This is different from traditional backup solutions, which create separate copies of the stored data. Because snapshots only need to copy the metadata or index at the time they are taken, they can be near instantaneous, have little performance impact and require little incremental space. IT organizations can take snapshot-based backups more frequently and improve recovery point objective. Vendors such as CommVault and analysts such as ESG have acknowledged the shift to snapshots as a viable option for backup and recovery.

However, it is important to note that not all snapshot implementations are created equal. Each of the implementations has different storage requirements and pose different restrictions on their use. The preferred implementation of snapshot is redirect-on-write (ROW). In this method, any updates to existing protected data are redirected to a new location. None of the existing data in snapshots needs to be copied or moved. As a result


ROW snapshots do not suffer the performance impact of the alternative copy-on-write snapshot implementations. The performance impact for copy-on-write snapshots limits their applicability for primary data.

Another consideration when implementing snapshots is the granularity of data that can be protected. This determines the space overhead of the snapshots taken. Smaller block sizes result in increased sharing of data between snapshots and greater space efficiency. With large blocks, a change to a small portion of a block would create a full new block with mostly duplicate data, causing the snapshot size to be much larger than the amount of data changed.

Figure 1: Snapshot granularity of traditional infrastructure can result in significant amount of redundant data

The last aspect that needs to be considered for snapshot design is the unit of data that can be protected and restored by the storage system. Traditional storage deployments typically operate at the storage object or volume/LUN level with little to no understanding of what is stored in those containers. In virtualized environment, this results in a simultaneous snapshot of tens-to-hundreds of VMs, each with varying change rates. Consequently, it puts the burden on the administrators to map the different VMs to the storage objects such as LUNs or volumes. This results in additional steps and greater system complexity, especially when recovering individual VMs. In the traditional approach, snapshot schedules can only be set at a LUN or a volume level, leading to practices such as creating one LUN per VM as a workaround in order to create individualized snapshot VM schedules.

An alternative to this method is taking a VM-centric approach to storage and data protection. In this scenario, storage understands and operates at the virtual disk or VM-level. So snapshots are taken at the VM-level and administrators can set schedules and retention periods at the VM-level to meet service levels. Recovery is simple as administrators can restore individual VMs without dealing with the underlying storage objects.

Snapshot granularity can impact storage efficiency over time


Snapshots on Nutanix

This brings us to the snapshot implementation on the Virtual Computing Platform. Nutanix OS implements redirect-on-write, VM-granular snapshots. When a snapshot of a VM is initially taken on the Nutanix Virtual Computing Platform, the system creates a read only zero-space clone of the metadata (i.e. index to data) and makes the underlying VM data immutable or read only; no VM data or virtual disks are actually copied or moved. The system creates a read only copy of the VM that can be accessed similar to its active counterpart. Nutanix snapshots take only a few seconds to create, eliminating application and VM backup windows.

After a snapshot is taken and as the VM continues to run, any updates to existing data and new writes are redirected. The original data in the snapshot remains unchanged and the unchanged data is shared across the snapshots and active VM. The Virtual Computing Platform handles this transparently so there is no change to how applications and the virtualization stack accesses the VM.

From an efficiency standpoint, Nutanix snapshots can be taken with byte-level resolution. This byte-incremental implementation means that only the changed data is captured between successive snapshots. For even greater efficiency, all the data stored on the Virtual Computing Platform including the snapshot can be compressed and deduplicated. Even though individual deployment savings will vary with the specific workloads, average deployments depending on the workload have seen anywhere from 25% to 75% reduction in the amount of space needed.

Figure 1: Nutanix snapshots are more efficient with byte-level granularity

The VM-granular snapshots can be set to be either crash consistent or VM-consistent and can be scheduled on an hourly, daily, weekly or monthly basis depending on the Recovery Point

Nutanix snapshots have

byte-level resolution


Objectives (RPO) and retention needs. The choice between taking crash-consistent or VM-consistent snapshots should be based on recovery needs. Crash consistent snapshots are instantaneous and are sufficient for workloads able to recover from operating system (OS) or VM crashes. Stateless applications such as web-servers are best protected through crash consistent snapshots. The alternative VM-consistent snapshots take advantage of host framework and services such as Microsoft Volume Shadow Copy Service (VSS) to quiesce the VM and supported applications; rendering them in to a known or consistent state. In the case of VMware running Microsoft Windows guests, VSS support is provided with VMware tools running in the guest OS. Using deep integration between Nutanix Virtual Computing Platform and VMware vSphere, the VMware tools are called to quiesce the OS and supported applications such as Microsoft Exchange and SQL Server before the Virtual Computing Platform takes a VM-consistent snapshot of the VM.

Additionally, multiple VMs can be grouped together in a Nutanix protection domain enabling them to be operated upon as a single entity with the same RPO. This is useful when trying to protect complex applications such as Microsoft SQL Server-based applications or Microsoft Exchange. The main advantage of using a protection domain approach of grouping VMs versus the traditional SAN approach of consolidating different VMs on to a single LUN is VM portability. VMs can be moved between different protection domains on a Nutanix Virtual Computing Platform without the need for any data to be moved or copied. For traditional SANs, changing a VM’s SLA will most likely require migrating the VM to another LUN or volume.

Because of the unique NDFS design leveraging a shared nothing distributed approach to metadata, there is no upper limit to the number of snapshots that can be taken with the Nutanix Virtual Computing Platform. This scalable approach eliminates the need for separate storage systems for backup and long term archiving, as the VM snapshots are stored across the entire cluster that makes up Nutanix Virtual Computing Platform.

Nutanix snapshot technology forms the basis of a unique set of functionality and ecosystem for high availability and disaster-recovery. The first feature that builds on the Nutanix snapshot capability is VM-granular cloning.

Keeping Data Optimized Nutanix Virtual Computing Platform runs a distributed data management service in the background. The MapReduce-based service is responsible for executing tasks such as metadata optimization, garbage collection of deleted VMs, data reduction, tiering, consistency checking, and rebalancing to optimize data across nodes and flash/disks with minimal impact to performance.


Cloning can be used for a variety of reasons including deployment and recovery. Integration with the virtualization stack with functionality such as VMware vStorage APIs for Array Integration (VAAI) and VMware View Composer API for Array Integration (VCAI) enables administrators to simplify VM deployment using integrated cloning. For the purpose of this document, the discussion will focus on recovering VMs.

The Virtual Computing Platform enables user-driven recovery of individual VMs from snapshots. This is done by either replacing the existing active VM with the snapshot copy or by creating a separate clone of a snapshot preserving the active VM. Depending on settings of snapshot, the recovered VM will either be crash-consistent or VM-consistent upon recovery.

If needed, administrators can create a clone of a Nutanix VM-granular snapshot for the purpose of recovering a single file without taking up additional space. Compared to a traditional LUN/volume based approach, a VM-granular snapshot approach eliminates the need for first recovering the storage object (LUN/volume) and then identifying and mounting the VM, and recovering the file.

Extending Snapshots over the Network: Backup and Disaster Recovery

Nutanix VM-granular snapshots also make it possible to efficiently replicate individual virtual machines from a primary Virtual Computing Platform to one or more secondary Nutanix clusters. By supporting a fan-out and fan-in or multi-way model for replication, the Virtual Computing Platform can create flexible multi-master virtualization environment for backup and disaster recovery. Deployments supporting numerous remote and branch offices can benefit from a flexible deployment model.

Figure 3: Multi-way protection domains make DR flexible


Since the software-defined replication functionality builds on VM-granular snapshots, policies for replication are also set at the individual protection domain level rather than working at the LUN/volume level. Only byte-level changes between snapshots of individual-VMs are sent over the network to the remote cluster. NDFS also enables another host other than the one serving IO on the active virtual disk in the cluster can do the work of calculating the changed blocks; eliminating performance bottlenecks for critical VMs and their corresponding hosts. So all nodes in the cluster participate in replication.

Figure 4: Eliminate bottlenecks by using all cluster resources for replication

To make the most out of WAN connectivity, the data can be deduplicated and compressed before it is sent over the wire. First the fingerprint of changed blocks for individual VMs are sent from the primary system to the different destinations. The different destination systems report back with the unique blocks they need to create the destination, which is sent back by the primary system. Deduplicating data over the wire can effectively cut the bandwidth required by as much as 90% versus host-based full-copy backup solutions.

Nutanix VM-granular replication also makes it possible to create an affordable disaster recovery solution. The converged compute and storage approach used by Virtual Computing Platform along with the VM-centric approach to replication makes creating a disaster recovery solution very simple. Using the protection domain concept, the groups of related VMs can be replicated together and those VMs can be brought up on the secondary site with a single command in case the primary site is down. Because the workloads are virtualized and replication is not hardware dependent, the secondary site can be different from the primary cluster. This is especially useful for remote sites deployment with centralized backup and disaster recovery.

The Future is RESTful

Nutanix Virtual Computing Platform provides an exhaustive list of REST APIs accessed through the Nutanix Prism management framework to various functions including around data protection and disaster recovery. These APIs can be explored through the Nutanix Prism

VM-granular Replication with Nutanix

Host-based or storage based replication


API explorer. The REST APIs are the foundation for the Nutanix Prism management interface and for failover run book automation through Nutanix Storage Replication Adapter (SRA) for VMware Site Recovery Management (SRM).

Figure 5: Nutanix Prism APIs enable integration with VMware SRM

In addition to delivering the functionality needed by VMware SRM for replication and VM management, using REST APIs the Nutanix SRA is able to tie directly to the user profile in Nutanix cluster administration for authentication and authorization. Nutanix Prism APIs can also be used to automate workflows using snapshots and replication for backup and disaster through scripting languages, or workflow engines. The Prism APIs are also used to create an automated run book for failover, automatically registering the VM at the DR site in VMware vCenter and powering them on. For example, a custom script can be created using the Prism APIs can trigger a Virtual Computing Platform to take and replicate a snapshot of the group of critical VMs making up an order-entry system, based on the number of transactions being executed.

Summary

With the increasing use of virtualization for critical workloads it is no longer optional to deploy data protection and disaster recovery. With VM-granular snapshots and recovery, use of policy-based protection domains, VM-granular replications, integration with the virtualization layer (ex: VMware SRM), and the use of REST APIs, Nutanix Virtual Computing Platform provides the functionality to backup critical data, protect applications, and survive disasters efficiently and cost effectively.

Protected VMs vCenter

SRM with Nutanix SRA

Controller

Primary Site Secondary Site

Protected VMs

vCenter

SRM with Nutanix SRA

Controller

Nutanix Replication

REST APIs REST APIs

Documents

Nutanix Data Protection and Disaster Recovery Tech Note_final.pdf