We’ve done disk-to-disk backup. It’s time we talked about flash-to-flash

With 1990 to 2010 being the era of disk-to-disk (D2D) backup, doesn’t it make sense – with the rise of all-flash arrays – to have flash-to-flash (F2F) backup?

One reaction is to say, “No. Flash is too expensive,” and there’s merit in that. Flash is more expensive than disk but there are countervailing points of view. TLC (3 bits/cell) and QLC (4 bits/cell) have brought the price of flash down and some IT sites need faster restoration than a D2D system can provide.

That was shown with the pioneering FlashBlade system from Pure Storage, which provided secondary, unstructured data storage on flash alongside Pure’s primary storage FlashArray. This has been followed by systems from StorONE (S1:Backup), VAST Data, IBM’s FlashSystem, and most recently Infinidat with its InfiniGuard system – five systems in all.

The whole aim is to ingest backup data faster and restore it faster than a D2D system such as Dell EMC’s market-leading PowerScale/DataDomain line and its cohort: ExaGrid, HPE StoreOnce, Quantum DXi and Veritas’s 5400 and 5300 appliance. The ExaGrid system is scale-out whereas the others are scale-up, hence each model in a line has capacity limits.

Most D2D backup appliances have a maximum bandwidth of around 100TB/hour:

  • Dell EMC PowerProtect – 94TB/hour
  • HPE StoreOnce – 104TB/hour
  • Quantum DXi – 99TB/hour

ExaGrid stands out through its scale-out design, with an 8-node system delivering up to 488TB/hour (61TB/hour per node).

The newer flash-era backup targets start in the 180TB+ per hour area with the scale-out VAST Data reaching 1.15PB/hour with an eight-node system.

Pure Storage

The Pure Storage FlashBlade was first introduced in March 2016.

In October 2018, Pure struck a deal with Veritas to have its NetBackup and CloudPoint backup software use FlashBlade as a target array. Veritas joined Commvault, Rubrik, and Veeam in doing this.

FlashBlade reached the billion-dollar sales level in 2021, a fairly conclusive demonstration that there is a market for all-flash unstructured data, secondary storage systems.

StorONE

StorONE’s S1:Backup system is a target for Commvault, HYCU, Rubrik, and Veeam’s data protection software and based on a 4–8 SSD flash tier with automated movement of older backups to more than 15PB of RAID-protected disk. The clustered S1 array can offer instant recovery by functioning as a standby production system.

Unfortunately, StorONE does not provide bandwidth numbers but we would expect throughput greater than 150GB/hour.

IBM 

The FlashSystem array has been proposed as a backup target for Veeam. Steve Lusnia, IBM North America FlashSystem Sales, said “Look at IBM FlashSystem as a best option Veeam can use as a backup target.”

The latest FlashSystem has 38.4TB QLC-based flash drives available with onboard dedupe and compression raising this to 116TB effective capacity, according to IBM. A FlashSystem 7300 can support up to 92 drives, meaning 45.5PB of effective capacity. That product has a 45GB/sec bandwidth, implying a 162TB/hour ingest and restore speed.

VAST Data

VAST Data set up a Data Protection division last year with George Axberg as the VP in charge. At the time VAST Data president Michael Wing said “His expertise and leadership at VAST will help drive the next major inflection point in data protection applications: the move from disk to all-flash architectures.”  

Axberg said “This is the biggest thing to happen to backup targets since they moved from tape to disk 20 years ago.” There is a partnership with Commvault to help things along and we would expect a flurry of other partnerships to follow, with backup software vendors such as Cohesity, HYCU, Rubrik, Veeam, and Veritas.

Infinidat

InfiniGuard software runs on an InfiniBox array and can store backup files from Commvault, IBM Spectrum Protect, Veeam, Veritas, and others, and from databases such as IBM DB2, Oracle, SAP, and SQL Server. 

Unlike the other vendors here, InfiniGuard is based on disk, which is accelerated to all-flash speed by DRAM caching.

Comment

So far there have been no indications from the legacy backup target system vendors about adding flash-based target arrays. We would be amazed if their development engineers were not going through what-if-we-added-flash exercises, though. A drive, after all, is just a drive, and primary storage arrays adopted flash – so why not backup target systems? Ransomware recovery would be so much faster from flash than from disk.

Our bet would be on Dell EMC bringing out a flash-based PowerProtect system and pitching it against Pure’s FlashBlade, which is probably the market leader in flash-based backup target systems.