New forms of direct-attached storage emerge to meet VM I/O demands

Feature

New forms of direct-attached storage emerge to meet VM I/O demands

Direct-attached storage used to be ubiquitous. Servers and storage were at one time synonymous, to the point that it became common practice that servers were replaced when they ran out of disk capacity. As data centres grew and storage became more expensive and strategically valuable, especially with the advent of server virtualisation, the case for sharing it became irresistible.

The result was the storage array, in SAN or NAS form, and DAS nearly vanished from enterprise storage. But while SAN and NAS have become the de facto standards in the data centre, alternative forms of arrays and a new breed of DAS are emerging.

The storage performance gap

Traditional forms of storage arrays work well for single-application servers, but with virtualisation generating more I/O per physical machine, access has become increasingly random. That’s made it harder for storage to match server performance.

Storage managers can buy more spindles to improve I/O, but this option brings with it cost, space, heat and power issues. Effectively, host performance has outstripped that of the array because the storage performance increase curve is less steep than Moore's Law.

Workarounds exist. Virsto Software has applications that plug into the hypervisor and perform the sorting required to sequentialise storage array I/O and so boost performance. Because this software is new and Virsto is a startup, there are still few IT organisations using this approach. The more common approach for better random I/O performance is to add solid-state drives (SSDs) to storage arrays.  

The new breed of direct-attached storage goes one step further. Bringing storage closer to the server alleviates the problems of expensive shared storage. It also means the storage system's resources -- including the network -- are dedicated to a single machine, which then enjoys better performance. Effectively, a single virtual machine (VM) host can now ensure that its storage runs at full utilisation, so why share that system and add to its workload?

Use cases for the new generation of DAS include highly dense installations such as private clouds and virtual desktop infrastructures, as well as those requiring access to large volumes of storage, such as images and video. It is used mostly by companies not yet ready to buy into the locked-in costs entailed in a traditional SAN.

Vendors and products

Buyers of DAS systems today are almost spoilt for choice. Taking a leaf out of Google's book is Nutanix. The company's Complete Cluster appliance is a clustered DAS system that consists of a scale-out compute and storage unit. This is integrated in a 2U block that contains four x86 server nodes, each with dual Intel Xeon 5640 processors, 320 GB of PCIe flash, 300 GB of SATA SSDs, 5 TB of SATA drives and from 48 GB to 192 GB of RAM. Each server includes a VMware hypervisor with a controller from Nutanix running in a VM. To scale, you can add servers numbering in the hundreds, according to the company.

Another example -- from VM6 -- combines direct-attached storage with software that brings disparate storage into a cluster. VM6 software installs on servers with Microsoft Hyper-V and turns the servers' disk drives into a logical pool of networked storage, mirroring it across the servers. VM6 says the software creates a virtual SAN that uses continuous replication to provide resilience.

Additionally, the company claims that by creating a virtual SAN, its software allows smaller businesses to repurpose existing servers and storage to help eliminate performance bottlenecks created by a random I/O load from virtualisation. The product runs on Windows and needs only a network connection between servers.

Storage vendor StoneFly has developed the Storage Concentrator Virtual Machine (SCVM) software, which integrates VM server hosts and storage in a single 2U, 3U or 4U box. SCVM is designed not just to simplify the integration of storage but also to boost storage performance. A single chassis delivers up to 108 TB arranged in multiple RAID configurations with SATA, SAS or SSD, tiered as required and expandable to 324 TB. The integrated servers access storage using from four to 11 1 Gbps iSCSI links.

Oracle's Database Machine is a more extreme example of clustered DAS in a box, and it’s an application-specific example of the company's Exalogic-branded systems, which integrate servers and storage. Consisting of 14 Linux or Solaris servers with 128 CPU cores, the 42U system is attached directly over InfiniBand to 5.3 TB of flash and up to 150 TB of disk storage. The system uses a flash cache for faster access to data and speeds up log writes using a battery-backed DRAM cache in the disk controller. Dedicated to database serving, the aim of the design is to boost performance of both SQL queries and storage I/O. Oracle claims its parallel storage access architecture can deliver up to 75 Gbps of I/O bandwidth and up to 75,000 database IOPS. The machine can be expanded by adding InfiniBand links and switches.

Fusion-io's ioTurbine goes further in integrating storage into the server. It uses PCIe-connected flash memory to deliver server-side caching for virtual machine hosts via a software layer that plugs into VMware's ESX hypervisor and represents its flash-based ioDrives as an available storage pool.  According to the company, front-ending a standard SAN or NAS in this way accelerates performance of virtualised workloads and I/O bottlenecks, which then enables more VMs per host, all without configuration changes for guest machines.

Virtual storage appliances to replace controller hardware?

Another method of bringing storage closer to the VM host is the virtual storage appliance (VSA), an example of which is Hewlett-Packard's LeftHand P4000 Virtual SAN Appliance Software. Run atop a hypervisor, the VSA can provide features more commonly found in controller hardware, including snapshots and replication, which can replace traditional backups. This is a big advantage when data volumes are so huge that backing up in the traditional manner would take days.

Today, software-based controllers cannot deliver the performance of dedicated hardware, but Moore's Law guarantees this will change as host processing power increases. As such appliances become able to handle larger volumes of data, host servers will share their storage between VMs.

With improved performance levels and no additional, expensive hardware to buy, VSAs coupled to a high-performance DAS could start to mark the decline of the traditional storage array. Instead of sharing storage between physical servers, the DAS will be shared by VMs running on networked hosts. This brings advantages of easier management and resilience. VMware's vSphere Storage Appliance does this (as does HP's P4000), allowing the benefits of server virtualisation to be used across two or three servers.

This will not be an overnight shift, but the growth of server virtualisation and its storage demands -- even in SMB environments -- seem likely to provide a growing market for the new breed of direct-attached storage.

Manek Dubash is a business and technology journalist with more than 25 years of experience.


Email Alerts

Register now to receive ComputerWeekly.com IT-related news, guides and more, delivered to your inbox.
By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

This was first published in November 2011

 

COMMENTS powered by Disqus  //  Commenting policy