Block storage: On-premise vs cloud

Block storage in the enterprise is usually provided by storage area network (SAN) hardware and most often associated with storage for databases, and enterprise applications built on database technology. But it is also used for storage in analytics and increasingly, machine learning and artificial intelligence (AI).

Virtual machines and containerised application are another core application, as is the provision of temporary, high-performance storage for tasks such as caching.

Block storage, however, is frequently demanding in terms of performance, especially IOPS and throughput. Over the past few years, cloud suppliers have invested heavily in technology to narrow the performance gap between cloud block storage and on-prem.

In some cases, cloud block storage now exceeds what local storage can do, but with some added advantages. These include ease of scaling and elasticity, lower costs or at least, capex savings. Cloud block storage also can also offer built-in support for dynamic tiering and security features such as immutable snapshots, and the ability to replicate files across multiple availability and geographical zones for resilience and recovery.

Microsoft Azure

Microsoft Azure offers block storage through Azure Managed Disks, and through Azure Elastic SAN, as well as container storage (for Azure Kubernetes Service) and, although it is technically object storage, through Page Blobs for Azure VMs.

Azure Managed Disks come in Ultra, Premium (v1 and V2) and Standard SSDs, and standard hard disks (HDDs) for less frequently accessed data. Microsoft claims sub-millisecond latency for Ultra, positioning it as the best option for “top tier” databases and transaction-heavy workloads, as well as SAP HANA. Ultra, however, trades performance for features and some services, such as zone redundant storage, require Premium or Standard disks instead.

Elastic SAN is aimed at environments that run multiple applications across a single storage pool, which Microsoft describes as “on-premises SAN capabilities in the cloud”. Azure claims millions of IOPS, and double-digit Gbps as well as built-in disaster recovery business continuity and threat detection. Elastic SAN is priced at £0.069 per GiB per month, the same as Premium SSD v2.

AWS

AWS’s primary block storage is Amazon Elastic Block Store (EBS). EBS is available in tiers, including general purpose SSDs, Provisioned IOPS SSDs, and throughput optimised and cold HDDs for lower-cost storage.

EBS can deliver up to 256,000 IOPS and 4,000 MBps of throughput, which AWS says is 4x that of general purpose volumes. AWS positions Amazon EBS io2 Block Express as suitable for HANA, Oracle, Microsoft SQL and DB2 databases, with general purpose volumes recommended for less demanding applications.

EBS also supports snapshots for backup and recovery, and data encryption in transit and at rest. AWS’s Instance Storage provides temporary, high performance block storage for buffers, caching or applications that run on EC2 instances that do not require persistent data.

EBS pricing ranges from $0.0928 per GB per month for General Purpose SSD (gp3) and $0.0058/provisioned IOPS per month (the first 3,000 are free) in the UK, to $0.0174 per GB per month of provisioned storage for Cold SSD. Although Amazon does not list EBS SAN options, it does describe its EBS and its virtual storage array as a “virtualised SAN in the cloud” in its technical literature.

Google Cloud Platform

Google Cloud’s block storage comes in three formats: Hyperdisk, Persistent Disk and Local SSD.

Google claims 500,000 IOPs and 10 GiB per second for its highest performance, Hyperdisk tier. Hyperdisk Pools allow Google Cloud customers to pool all their storage to a single pool, which the hyperscaler claims reduces the total cost of ownership by up to 40%. Hyperdisk allows users to provision IOPS separately from capacity, making it a lower-cost option than its other storage options.

The technology also includes Instant Snapshots for backup, DR and resilience. Hyperdisk itself is built on Google’s Titanium technology, custom silicon that “offloads processing from the host hardware”.

Hyperdisk itself comes in Throughput, Balanced, and Extreme versions as well as Storage Pools; Google recommends Hyperdisk for SAP HANA and SQL Server workloads. Hyperdisk, Google says, offers 7.5x more write throughput and 2.5x more read throughput than Persistent Disk, and at a lower cost.

For low-latency, high performance temporary storage Google also offers Local SSD, and Persistent Disk, for virtual machines.

IBM

IBM’s Cloud Block Storage for VPC is the supplier’s primary block storage offering for its cloud virtual and bare metal servers. Recommended uses include databases, enterprise applications and applications with high transaction workloads.

IBM Cloud Block Storage for Classic provides storage for bare metal servers or containerised applications on the “classic” IBM cloud.

Cloud Block Storage supports dynamic data allocation, and allows users to move storage capacity between servers based on workload requirements. Block Storage for VPC and classic is available on hourly and monthly pricing. VPC has three IOPS tiers, 3 IOPS/GB, 5 IOPS/GB and 10 IOPS/GB. The latter is aimed at the most critical workloads, including NoSQL, analytics, and AI and machine learning.

Through block storage on VPC, IBM also supports snapshots for backup and security, and scaling across availability zones.