The Role of Flash in the Datacenter

All-flash storage supercharges application performance, but the optimum way for IT teams to deploy flash is still a subject of debate. This the first in a series of blog posts exploring the advantages of all-flash hyperconverged infrastructure for enterprise apps.

Flash is taking the datacenter by storm. However, many IT teams are still uncertain how to deploy flash to achieve optimum performance for critical enterprise apps—especially without compromising other important IT objectives. Is traditional network storage the right approach, or is it better to deploy flash inside your servers?

The IDC report discusses Optimizing Flash for the Most Demanding Application Requirements how all-flash hyperconverged infrastructure (HCI) with the right underlying architecture is the best option, noting that today’s applications are increasingly designed for co-located compute and storage.

Shortening the I/O Path Length
At the core of IDC’s argument is this:

Hyperconverged systems shorten the I/O path from the application to the storage, resulting in significantly lower latencies than network storage can deliver.

It’s pretty hard to argue with this statement. Think about the path that an I/O read request follows in a traditional three-tier infrastructure (server, storage network, storage):

Application→Kernel→PCI Bus→NIC or HBA→Across storage network→Storage NIC or HBA→Storage kernel→PCIbus→drive controller→Device

Versus the path that an I/O request follows using HCI based on web-scale architecture with intelligent tiering:

Application→Kernel→PCI Bus→Drive controller→Device



Today, given the low latency delivered by a single solid-state drive (SSD), the network is the slowest element in the I/O path. And since a relative handful of devices can deliver extremely high IOPS and significant capacity, it’s pretty hard to justify the cost and complexity of a storage area network (SAN). Putting your flash storage across a SAN is like putting wooden wheels on an electric car.

Why Hyperconverged Infrastructure?

If you accept the conclusion of the previous section, then it might still occur to you to wonder what the advantage of HCI is versus simply installing SSDs in your existing servers? The answer to that question comes down to data management and data protection. If you’ve got tens or hundreds of servers each with several SSDs, at some point, one of those servers is going to go offline or fail.

One of the benefits of a correctly architected HCI solution, as IDC highlighted, is a distributed file system that’s self-managing and self-healing. If a drive or a server fails, recovery is fast and all data remains accessible.

For example, the web-scale Acropolis Distributed Storage Fabric (DSF) pools local flash storage across server nodes. Now all flash performance doesn’t require an all flash array. Storage is provisioned and managed at the VM level, with data locality managed automatically. Move a VM and its data moves with it. DSF simplifies data management while protecting the availability of critical applications and data. DSF also extends all flash performance to select VMs on hybrid SSD+HDD nodes with the VM Flash Pinning feature.

Data Reduction

You probably already know that flash comes at a premium to spinning disk. (The cost of flash in terms of $/IOPS is much lower.) In its report, IDC also noted that data reduction technologies such as inline deduplication and compression are highly desirable for all-flash HCI. These technologies increase the effective capacity of your flash storage pool, and also increases flash endurance by reducing total write activity.

In addition to inline compression and deduplication, Nutanix DSF also includes thin provisioning and space-efficient snapshots and cloning, allowing you to store more data with less total storage and making Nutanix HCI deployments extremely cost effective.

Beyond Performance

As important as performance is to enterprise applications, you shouldn’t have to compromise other IT objectives for it. Data protection and disaster recovery are obviously critical, and the overall manageability of your IT environment is important too. IT teams need to spend less time dealing with infrastructure and more time enhancing business value.

A separate all-flash array is another silo of infrastructure to be managed along with whatever you currently have. Data protection and DR methods are certain to be different and possibly unproven.

In the next post in this series, we’ll look more closely at how hyperconverged infrastructure solutions from Nutanix unlock the power of all-flash performance while delivering superior availability and manageability.

Still looking at a standalone all flash storage solution from {EMC-HP-PURE-NETAPP-IBM}? Talk to Nutanix first to find out why flash storage for your enterprise apps shouldn’t live on a network.