May
31

All-Flash Converged Storage

All-Flash Converged Storage

May 31
By

hand holding an iphoneThe introduction of the smartphone enabled customers to carry a single device that can run multiple applications, instead of carrying multiple devices that each work for one app (such as a music player, camera, calculator, phone, etc.). Similarly, it is time to collapse the islands of storage that exist in a company by having the ability to run multiple workloads from a single all-flash storage backend.

When all-flash storage was introduced to the market, the use cases for such storage were primarily high transaction workloads with block protocols. While this was a great use case to address with flash back in the day when cost of flash was 100x of hard drives and the capacity of flash was limited, the current trends in flash capacities especially with the advent of MLC and TLC flash memory technology and trending down in prices allow all-flash to become a true converged multi use case platform.

The following chart indicates the costing trends of NAND flash and hard drives. It shows the crossover of SSD and HDD pricing happening soon.

10-year technology cost/ terabyte projections 2014-2023

Figure 1 Source

The areal density of SSDs have also surpassed the densities of HDDs in recent times, driven by design and manufacturing advances such as 3D NAND. As an example, there was a recent announcement of 100TB in a single 3.5-inch 3D NAND SATA SSD!

The above two trends of decreasing cost and increasing density makes all-flash storage more conducive for a broader set of workloads. The digital transformation that’s happening and is being widely adopted in enterprises has resulted in a considerable amount of growth in unstructured data. For example, file services, videos, technical applications, archiving, backups, etc. comprise a growing amount of the data.

Backup workloads, which contribute to the growth of unstructured data, have also evolved in functionality over the last few years. Many backup applications now provide a zero-time recovery window by providing the ability to spin up a virtual machine from backups directly instead of having to restore the data first. This feature offers a significant advantage for customers during outages. The ability to instantly restore is dependent on the storage system being able to deliver sufficient IOPS and low latency, where flash is a much better fit. Another reason for all-flash storage in the backup space is the continued need to reduce the backup windows for enterprises.

In addition to unstructured data, organizations typically have a preponderance of virtual machines with lower transactional processing needs with guest VMs running service applications or data serving applications with mid to lower tier databases. Virtual Desktops Infrastructure (VDI), which required IOPS and latency that HDD solutions cannot meet, is seeing increased adoption in many of the mid-market segments as well.

The needs of a converged workload solution are currently unmet because either storage systems do not have the right feature set, or they are too expensive. The design of such a storage system needs to have certain properties to support the newer  workloads:

  • Scale with data growth
  • High availability
  • Data integrity and durability
  • Security
  • Disaster recovery
  • Economy in cost
a model of OneBlox 5120

The StorageCraft OneBlox all-flash appliance was built with this kind of feature set in mind from the beginning. Some of the architectural design points that enable this feature set are explained in the following text.

Scale with data growth

Scalability cannot be a bolt-on feature for a storage system. It needs to be designed-in from the start. There are two aspects in providing scalability: One is the storage system architecture allowing scale out of nodes starting out from a single node and second is the ease of management of the growth in data.

OneBlox Scale-out architecture: The OneBlox architecture provides file access protocols with a peer to peer  distributed node architecture based on key value stores and object storage. This architecture provides the ability to pool storage capacity across the systems while paralleling I/O across all the drives in the cluster without any single point bottlenecks in the data path.

Management Simplicity: Ease of clustering and self service of administrative tasks has been a design goal of the OneBlox architecture from the beginning as well. Using platform services such as zero config and auto discovery, cluster formation is intuitive and  simple without requiring professional services. Cloud based management service gives insight into analytics, telemetry and topology of all the devices from a single dashboard.

High Availability

In a large scale cluster system, the hardware components will be in large numbers and the storage system software needs to be designed in a way to handle failures  graciously. On a node or drive failure, OneBlox software rebalances the objects to bring the data back to a healthy stae. and migrates the file system services to provide continuous ability to applications or VMs.

Data Integrity and Durability

Online Research Concept on Data Around the World

Data Integrity is essential to guarantee no loss or corruption of users’ stored data. Data Durability refers to the aspect of a storage system that ensures data is persisted when acknowledging writes to the application.

Data Integrity is ensured through many mechanisms in the OneBlox system:

Cryptographic hashes of the content along with check-summing of the data checks for integrity when data is read or written. Data is also scrubbed periodically when various background activities are performed on the system. When an object is detected to be corrupted, the object is recovered from the additional copies stored in the cluster. Parity bits are stored as part of the disk layout to provide a higher level of data integrity when compared to the TLC and MLC flash uncorrectable bit error rates.

Data is made durable in the face of power failures by using NVDIMMs to log the data in non-volatile memory before committing to the backing store, as well as replicating the log to other nodes in the cluster before acknowledging the writes to the application. When the node comes back up after restoring power, the intent log on NVDIMM is replayed, or in case of a node takeover of file system services in a cluster, the replicated log is replayed. Data reduction techniques improve the durability of flash. These techniques benefit endurance lifetimes of the media by reducing the drive writes per day. Log-structured writes allow large sequential writes, thus avoiding write amplification from the user I/O pattern that consists of small writes or random writes.

Security

Although the OneBlox cluster may be located behind a firewall in a data center, security is still of concern for enterprises. There are various mechanisms implemented in OneBlox to provide additional levels of security. These mechanisms include:

  • Encryption for data at rest as well as over the wire for remote replication
  • Device Certificates to avoid man-in-the-middle attacks
  • Trusted Platform Module use to stop unauthorized apps from running on the device
  • https protocol for communicating to management service
  • Secure Boot to ensure booting only with trusted software
  • Constant monitoring and integration of security patches for vulnerabilities

The other core design aspect of the OneBlox software architecture is the immutability of objects. Even when data is overwritten, new objects are written to storage while leaving the old objects as part of a snapshot. This aspect of the design protects against attacks such as ransomware by enabling near continuous snapshots of the data. Customers can retrieve the copy of the data from an older snapshot just minutes before the attack.

Disaster Recovery

WAN-optimized object-based bi-directional remote replication in OneBlox software provides disaster recovery protection. Through the use of StorageCraft cloud services, incremental backups can be stored on the cloud as a means of protection against disaster as well.

Economical Cost

What makes the all-flash OneBlox solution economical is the pay-as-you-grow scale-out storage architecture in which each node requires nominal hardware resources as compared to scale-up architecture. This facet, combined with a bring-your-own-drive business model, lends the overall solution to be affordable for a much broader set of use cases.

Summary

The cost and density curve trending of flash relative to hard drives now makes it possible for flash based solutions to serve a larger variety of workloads for enterprises. A storage system with the right design properties at the right price point can serve these workloads. The OneBlox all-flash solution was designed to have these properties and enables customers to use this solution in wider ranging workloads.