Thursday, November 19, 2009

George Crump, Senior Analyst

Adding shared storage to the virtualization project ‘supercharges’ its capabilities. A virtual infrastructure built on a shared storage foundation allows capabilities like live machine migration, live storage migration, virtual host high availability, dynamic load balancing of virtual machines and off-host backup of the virtual environment. Going further, it is a key building block to a disaster recovery strategy and greatly simplifies overall storage management.

VMware has stated in the past that for every dollar spent on its server virtualization products the typical user spends five dollars or more on the underlying storage using traditional shared storage options. Before discussing the power of shared storage in a virtual environment, the manager of the SMB data center needs to understand that they can afford to purchase, implement and operate shared storage. This is because recent changes in technologies have brought down the price point of shared storage within the reach of the SMB budget and improved its ease of use.

The first step is to drive out the upfront costs. Shared storage in the enterprise has typically had a high cost because it meant purchasing proprietary storage from a single vendor and implementing that shared storage on a new network infrastructure like Fibre Channel. While the costs of these systems and their infrastructures has come down, the purchase of something new will typically raise the cost of a project. Second, because the storage is from a single vendor there is often, (always?), a premium price associated with it.

For the SMB it would be more cost-effective if they could leverage the investment they already made in servers, storage and network. A way to accomplish this is to use separate storage software from the storage hardware. All storage systems, whether they are designed for small business or large enterprise, have essentially two components; hardware and software. The software is the intelligence that manages sharing of the storage and other capabilities like replication or snapshots.

Companies like StarWind Software have, by separating the SAN “firmware” from the hardware, enabled the SMB to convert new servers with enough hard disk capacity into a fully-functional, fault-tolerant SAN. The “storage virtualization” software essentially turns any Windows server into shared storage for just the cost of a software application. Expansion of the storage can then be done with off-the-shelf storage hardware available from a variety of sources, driving down future storage costs as well.

Storage as an application especially makes sense to deploy in server virtualization projects were it is likely that servers are going to be consolidated. With storage virtualization software, the SMB can select any server, load it up with memory and disk drives and very quickly have a solid-performing storage server for the foundation of a server virtualization project. Since these are standard Intel based Windows servers, there is a high degree of interchangeability in components, like hard drives and memory. Also, software like StarWind's allows the SMB to build these in pairs, so high availability and redundancy can be achieved. The pairs however, do not need to be identical. Simply select two high-quality servers and the shared storage platform or the SAN is all set.

Another key aspect in making shared storage affordable for the SMB Virtual Server project is to similarly leverage the existing network infrastructure instead of implementing a new one. Most SMBs will have some form of IP network in place long before they begin a virtualized server project. As a result, a storage system that can be shared with the existing IP networks would be able to further keep storage costs down even if the storage network can use its own switch or networking gear to ensure high performance of the storage traffic.

There are two storage protocols that leverage an existing IP infrastructure. The first is the file sharing protocol NFS, which is commonly found in NAS environments (file-level storage). NFS , while supported by some server virtualization platforms, is not supported by all. If it were to be used, for most SMBs, the purchase of a specific NAS based appliance would be required. For hosting virtual machine images a fairly powerful, and even more expensive, NAS is also required. Additionally the hosts in the virtual infrastructure cannot boot from the shared storage, so NFS may require some local storage. Finally, any type of application, such as clustering, that requires traditional SCSI block storage will not work with NFS based storage.

The second protocol that leverages the existing IP network and is used for block-level storage (i.e. SAN) is iSCSI. iSCSI encapsulates standard SCSI commands and sends them across an IP network. This means that almost anything that can be done with a local SCSI hard disk can also be done shared, across a network with iSCSI, including clustering and booting from the environment. For the small-to-medium-sized business this may be the ideal situation. iSCSI allows the SMB to leverage the existing IP network that they already have.

The throughput of 1Gb iSCSI for most SMBs should deliver adequate performance with no significant network tuning. Since most, if not all, servers and networks today are already running at 1Gb speeds there should be virtually no change in the environment required. The only upgrade that the SMB may want to consider is adding a separate switch and 1GbE cards in the servers just for the storage traffic. Although it’s not required, it may help maintain a more predictable level of performance, especially during system backups and other high network load times.

Servers that are found in most SMB data centers have plenty of compute power available to them to use software-initiated iSCSI. This is the driver that does the translation between local storage commands and iSCSI. In the past it was suggested that iSCSI be implemented with specialized iSCSI cards, which did this translation on the card, freeing the server CPU for other tasks. While in some environments this may be needed, for most SMBs the server CPU can handle both tasks, running the virtual environment and doing the SCSI-to-iSCSI translation with little impact on overall performance. The key advantage of using software-initiated iSCSI is that standard 1GbE cards can be used in the servers, which are 1/8 the price of specialized iSCSI cards.

Once the storage platform is built, the SMB has to be confident that they can implement it and operate the environment. Since this iSCSI environment is built on a networking protocol that they are likely to be very familiar with, it is as simple as continuing to manage that network. Other than centralizing the storage and making it shareable, nothing new has been added. The SMB simply has another component on their existing IP network. In the StarWind case, as we pointed out in our review of the product [embed link to review], the GUI-driven interface of the product makes implementation and management of the environment very simple and straightforward.

The bottom line is that many SMB data centers can implement shared storage by simply adding software. And companies like StarWind make that even easier by having the software downloadable from a web site. The storage in the server that will act as the host for that software can also be repurposed as part of the server virtualization project. Essentially, the SMB powered by an iSCSI software solution can get shared storage and all the benefits that it brings to the environment with an absolute minimum in capital outlay. The case could be made that shared storage is more within the grasp of the SMB than it is, even for the large corporate enterprise.