Virtual Data Rooms

What networking, I/O interfaces, and protocols to use for a virtual data center, including virtual server and storage scenarios, will depend on specific needs and QoS requirements. Other factors include the type of I/O profile—large or small, random or sequential reads or writes—and the number of VMs and type of device adapters.  Depending on the version of a virtual infrastructure, such as VMware or Microsoft, some advanced features are supported only with certain protocols. For example, a VM may be able to boot off of any network interface or protocol, while the virtualization hypervisor may have restrictions or specific configuration requirements.

Another consideration is virtual desktop infrastructure (VDI) boot requirements  that result in a spike in network activity that would have otherwise been handled by a local disk device. On one hand, VDIs shift their disk I/Os to a server that in turn sends the data over a network to be used or accessed on the thin client or desktop. On the  other hand, placing the applications and data on disks attached to a server shifts the focus or burden of backing up desktops and subsequent networking demands. 

However, trading network resources to support backup rather than enabling a VDI is not an even trade, because backups tend to be large sequential streams focused on bandwidth, while VDI traffic may be smaller, burstier IOP, frame, or packet per second with a focus on low latency. Exercise caution with VDIs to avoid simply moving a problem or causing a new one elsewhere instead of indentifying and eliminating bottlenecks to improve QoS and customer service experience

While some VMs can use and leverage local dedicated and nonshared storage, most features for scaling and resiliency, including HA, BC, and DR, require some form of shared storage. Shared storage includes dual or multiported SAS storage arrays attached to two or more servers, iSCSI and Fibre Channel block storage, and NFS NAS storage. Other features, including clustering, dynamic VM movement or relocation, and server-free backup, will vary depending on version and type of I/O interface and protocol being used.

Performance will vary depending on specific VM configuration, underlying hardware architecture, guest operating system and drivers, and storage system configuration. Virtualization vendors have configuration guides and on-line forums covering various configurations and supported options, as do server and storage vendors. There is debate in the storage community among iSCSI, Fibre Channel, and NFS NAS vendors as to which protocol is the most efficient and best to use. Not surprisingly, some vendors claim their protocol and interface is the best. Others take a more neutral and consultative approach by considering what is the best tool and technique for the task at hand, along with individual preferences and existing or planned technology decisions. 

Other applications, such as Microsoft Exchange, can also have a requirement of running on block-based storage or on a Microsoft file system. Traditional databases such as Oracle, Microsoft SQL, or IBM DB2/UDB have had a preference and a recommenda tion, if not a requirement, to run on block-based storage. Databases such as Oracle can and do run on NFS storage and, by leveraging a feature in NFS V3 or later called direct I/O (DIO), can perform blocklike access of NAS-based storage without having to read or write an entire file. These rules and requirements are changing, and it is important to consult with vendors or product specialists as to specific guidelines and recommendations.

Networks and I/O connectivity are important and, with continuing convergence, the lines between storage interfaces and networks are becoming blurred. Networks and I/O involve networking, server, storage, and data professionals to coordinate the various interdependencies across the different technology domains. For example, if a virtual  server is moved from a VM on one physical server to another VM on a different physical  server, unless NPIV is being used, SAN personnel have to make zoning and configuration changes, as do server and storage personnel. Various tasks and activities are required to keep I/O and network capabilities in good order to support virtual data centers. 

Metrics for networking include:

• Bandwidth of data moved

• Latency and response time

• Frames or packets per second

• Availability and quality of service

• Top talkers between source and destinations

• Error counts, including retransmissions

Not to be forgotten or diminished in their importance are test and diagnostic devices. These include protocol analyzers, performance probes, network sniffers, and fault and error detection equipment. Some other examples include network and physical cable monitoring and diagnostic devices and workload generators. Some of these devices are protocol-based, while others work at the networking interface layer and still others at the physical cable level. These tools can be integrated and complemented by additional management software tools as part of an overall storage and storage networking management solution set.

Testing and diagnostic equipment can be utilized for design and development of components as well as testing of components individually and as part of an overall storage network. For example, a component may work fine by itself and with other devices under certain conditions, but when additional devices and workload are added, things may change. Unit and component testing for verification and validation and stress testing of an entire system under various workloads may uncover issues not seen in normal testing, such as impacts on memory or memory and network leaks. The combination of various equipment, operating systems, patches, and device drivers can add to the complexity of testing and diagnostics.