Data everywhere is growing at unprecedented rates. The increase in the value and volume of data has triggered the adoption of all-flash in enterprises worldwide. Yet many backup environments remain unchanged and unable to meet the new demanding recovery requirements, demands of cloud services, and the data reusability needs of modern use cases.  

A fast, reliable, simple, and flash-enabled form of data backup and recovery—at scale—is more critical than ever.  

Technology leaders are looking for better answers to the challenges associated with the recovery demands of data and the complexities they create for infrastructure. 

Pure Storage® and Cohesity forged a strategic partnership and brought to market a first-of-its-kind solution to address the challenges of data protection. Pure FlashRecover™, Powered by Cohesity® is an integrated modern all-flash data protection solution for rapid recovery at scale. The jointly developed solution is simple and fast and provides the most reliable backup and recovery for enterprise data. The solution allows organizations to do more with their backup data, such as restoring petabytes of data in hours—not days or weeks.

With storage innovation from Pure and modern backup from Cohesity, the integrated solution delivers: 

  • High performance  
  • Simplified management 
  • Scalable and efficient data reuse capabilities 
  • Streamlined support and procurement for modern data protection 
  • Recovery at scale  

With this integration, you can achieve up to three times faster backup and restore throughput than disk-based alternatives. This solution enables enterprises to perform full recovery of thousands of VMs and petabytes of data in just a couple of hours rather than days or weeks.

Solutions Architecture

To illustrate the capability to restore up to 1PB/day, we formed a 12-node FlashRecover cluster with a 15-blade FlashBlade® chassis. The source virtual machines were created on 27 ESX hosts configured with seven different FlashArray™ devices, as shown in the architecture below. 

Figure 1: Solutions architecture overview for rapid restore at up to 1PB/day.

You can host the source applications (VMware in this instance) on any storage. But for fast backup and fast restore, it’s imperative to host it on an all-flash array like FlashArray. Twenty-seven ESX servers were installed with VMware ESXi 6.7 and set up for ingestion virtual machines. To evenly balance the load, four ESX servers were mapped to one FlashArray. The ESX servers were connected to FlashArray over iSCSI. Datastores were created from seven FlashArray devices on which six VMDKs are created for every virtual machine. The ingestion virtual machines in the testbed were selected equally from ESX hosts.

The test illustrates the performance for 1PB of data per day using Linux virtual machines. Every ESX server was equipped with a 10 Gbps network interface, which enables around one Gbps of bandwidth between the ESX servers, FlashRecover, and FlashArray. To show higher and scalable bandwidth, multiple virtual machines were backed up and restored across ESX servers. Virtual machines are distributed evenly across the ESXi hosts and datastores. A total of 50 Linux virtual machines of various sizes are created on these ESX servers and distributed as equally as possible. The virtual machines are configured with six virtual disks, each with 350GB of logical size, with a total logical size of 2.1TB per virtual machine. Out of 50 VMs, 15 VMs are populated with 2TB of data and the remaining 35 VMs are populated with 1.75TB of data. Each virtual machine in the test consists of at least 105GB of unique non-dedupe data, with a total size of 91TB for backup and restore.

A full backup of the policy with 50 virtual machines was triggered and captured the performance characteristics and the total duration to complete the job, as shown in the screenshot below. A total size of 91TB is backed up across 50 virtual machines. The total duration to finish the backup job was about 133 minutes (about 2 hours and 13 minutes), which is a backup rate of about 11.4GB/s. 

Figure 2: Full backup job run with a data read of 91TB.

Rapid Restore at up to 1PB/day

To demonstrate the rapid restore of up to 1PB/day with FlashRecover, a full restore of the previous backup policy of 50 virtual machines with a total size of 91TB was triggered from the recovery tab of the FlashRecover GUI. A full copy recovery of 91TB across 50 virtual machines is restored to the original location. A peak throughput of 15.8GB/s is observed, as shown in the screenshot below. An overall run rate of 12+GB/s of throughput was achieved, which is equivalent to restoring 1+PB of data in 24 hours.

Figure 3:  Performance characteristics of restore.

The total duration to restore 50 virtual machines is 206 minutes (about 2 hours and 6 minutes). The total duration taken to finish the restore is tagged as the timestamp of the last restored virtual machine. And it doesn’t necessarily mean that all of the virtual machines took the same amount of time to finish. The figure below shows the details of the job run with the total duration for the restore job to complete and the restore timestamps for a few virtual machines. 

Figure 4: Restore duration of the recovery job.

The simplicity and high-performance linear scalable capabilities of FlashRecover enable you to achieve 1+PB/day of restore rates without adding much complexity. 

Explore the details of the architecture and restore performance of up to 1PB/day with an overview of FlashRecover and the solutions architecture.