Two pitfalls for enterprises implementing server virtualization

Virtualization is viewed as a panacea for many enterprise IT problems. From enhanced application availability to simplified disaster recovery to reduced infrastructure and costs, virtualization seems to provide all the answers. Virtualization also appears to offer simplified IT management and even "greener" computing solutions.

However, to get the most benefit from server virtualization, it is important that other elements of the infrastructure (particularly storage) compensate for the shortcomings of the virtualized environment. Otherwise, many errors will occur. Applications can unexpectedly slow to a crawl. What was supposed to be a cost-reducing computing alternative requires significant investment to realize its full functionality. Using virtualization to improve application and server uptime suddenly exposed painful weaknesses in other aspects of the IT infrastructure. Here are two of the most common pitfalls when enterprises implement server virtualization .

Pitfall 1: Choosing the wrong storage platform

One of the major benefits of server virtualization is the ability to move client applications in use between different server hypervisors. Whether this is done for orchestration planning, load balancing or disaster recovery, hardware independence is one of the key enablers for any virtualization implementation. However, if your storage is tied to specific server hardware, mobile applications can become complex or even confusing.

Network-attached storage is often used as a way to simplify virtual server storage provisioning. Network-attached storage capacity is very simple to set up, and capacity expansion does not require the involvement of a hypervisor. Unfortunately, there are performance weaknesses with using network-attached storage. Many applications (such as Microsoft Exchange) simply won't run using network-attached storage. For these reasons, most virtualization vendors will recommend SANs to those looking for more efficient application performance.

Fiber Channel Storage Area Network

With Fiber Channel SANs, users not only need to justify the added costs of Fiber Channel storage, switching and management, but they also need to configure expensive host bus adapters for every server they connect to the SAN. Those enterprises adopting existing Fiber Channel storage area networks will encounter few obstacles. To reap the significant benefits of server virtualization, this complete Fiber Channel infrastructure (including switches and host bus adapters) needs to support the NPIV (N-Port ID Virtualization) protocol. Most products currently do not contain NPIV.

Even with NPIV, VMware can only transfer guest programs between machines within a Fiber Channel zone. This means that although you have achieved hardware independence on the server side, all the physical servers in a group that can deliver client applications to each other are not dependent on a single Fiber Channel zone for storage (usually be an array or even a hard drive). Hardware independence on the server side can create dangerous multi-application hardware dependencies on the storage side.

Optimize storage solutions for virtualized environments

iSCSI (Internet Small Computer System Interface) or IP SAN (IP Storage Area Network) provide the best storage solution in a virtualized server environment, not only from the obvious cost advantage, but also from the availability of the virtual architecture, It is also the best in terms of flexibility and scalability. An iSCSI SAN storage system can also provide significant advantages to enterprises using WAN disaster recovery virtualization. Snapshots can also be used at the storage level to replicate data to a local or remote backup site.

In addition, iSCSI storage LAN storage systems have obvious WAN advantages over Fiber Channel storage LANs. Fiber Channel storage WAN replication requires the purchase of expensive FCIP (Fibre Channel over IP) gateways. WAN replication for iSCSI storage LAN storage does not require the purchase, implementation, operation and management of additional systems. iSCSI is a TCP/IP protocol that works natively across the WAN. Both Fiber Channel and iSCSI WAN replication can cause throughput degradation or packet loss over long distances. WAN or TCP/IP optimized devices for iSCSI storage LAN storage can alleviate this problem. This WAN or TCP/IP optimization device has no or little impact on FCIP gateways.

Trap 2: The over-provisioning dilemma

Even with the right storage area network solution, migrating applications to a virtualized environment can sometimes slow to a crawl. If the server hardware configuration is correct, the administrator cannot explain the cause. In this case, storage is usually the cause of the problem.

The efficiencies that virtualization brings to the infrastructure are achieved through deliberate over-provisioning using a hypervisor. Virtual guest applications are allocated a sub-optimal share of physical resources. This is done based on the principle that it is statistically impossible for all applications to require resources at the same time. The principle of proportional use is generally feasible in practice. However, most SANs and SAN storage are already over-provisioned, and the results of double over-provisioning of physical storage resources are catastrophic.

Since the storage infrastructure was really stretched, conflicts became an issue, and bottlenecks and buffer overflows occurred. Further complicating things for administrators is that these conflict issues can occur at multiple levels of the storage infrastructure.

At the level of an individual drive, the queue of input/output requests will grow. This problem will be more prominent when configuring slower SATA hard drives. In SATA hard drives, the queue depth is generally 0 to 32 requests, while in SAS (Serial Attached SCSI) or Fiber Channel hard drives, the queue depth is 256 to 512 requests. This means that enterprises looking to implement virtualized infrastructure need a SAN solution that doesn't limit their choice of backend drives.

At the storage LUN (logical unit number) layer, the hypervisor itself generally divides a physical storage pool or LUN into multiple virtual LUNs. These LUNs are then allocated to different virtual guest applications. These physical LUNs cannot differentiate between these client applications. Excessive resource conflicts can reduce storage performance.

Likewise, overprovisioning at the hypervisor level can cause problems at the SAN infrastructure level with host bus adapters, initiators, ports, and switches. These resources are often over-provisioned 8:1 or exceed the configuration of the storage area network itself. The compounding effect of this double over-provisioning not only degrades performance, but also causes request timeouts and application crashes.

Resolving Excessive Conflicts Using Virtual Storage Area Network Storage

One option is to turn off storage virtualization in the hypervisor and manually allocate LUNs to each guest application. However, many vendors do not support this. Doing so also loses critical virtualization capabilities.

Another option is to address the issue from the storage side, reducing local over-provisioning levels in the SAN architecture. Using a physical SAN, this is complex and will significantly reduce the efficiency of the SAN as a virtualization host. With virtualized SAN storage, this reconfiguration is not only simpler, but often enables the hypervisor to be treated differently based on the physical host to optimize overall SAN efficiency.

Indeed, a virtualized storage area network can also be used to spread a single LUN across multiple storage resources, further mitigating resource conflicts. Virtualized storage area networks provide the performance of storage area networks with the simplicity of network-attached storage.