april 3

Feng shui IT
Architectural balance trumps CPU cycles and throughput
Computerworld Opinion by GlassHouse CTO, Jim Damoulakis

http://www.computerworld.com/action/article.do?command=viewArticleBasic&articleId=9015458

Effective infrastructure design has always involved weighing an ever-changing progression of trade-offs in order to strike an acceptable balance among performance constraints (CPU cycles, bandwidth, memory/storage capacity) and cost based on technology choices available at a given point in time. Given today's reality of fast servers with multicore processors, memory capacity exceeding what once were disk capacities, and a seemingly infinite well of storage capacity, it's easy to lose sight of the fundamental principle of maintaining appropriate architectural balance.

Since it has become relatively easy to throw more technology at a solution, we now face the somewhat paradoxical situation where design flaws can occur because components are either too cheap (overdeployed) or too fast (inducing greater bottlenecks). The result can be sloppy configurations that are out of balance. Here are two current design challenges to illustrate the point:

1. Server virtualization and iSCSI

The relative affordability of servers has led to data centers filled with servers at sub-15% CPU utilization levels. The current popularity of server consolidation through virtualization stems in large part from its ability to improve efficiency by leveraging those idle processing cycles. At the same time, in the storage arena a growing number of environments are embracing iSCSI for storage connectivity. One key factor in the appeal of iSCSI is the lower cost of storage networking infrastructure because of its ability to leverage standard networking interfaces and switches. Unfortunately this indirectly depends on the availability of CPU cycles to handle the additional IP networking overhead on the server.

Combining the two -- virtualization and iSCSI -- seems like a no-brainer but can potentially result in a CPU bottleneck. One obvious solution, use of iSCSI hardware initiators, may or may not be a viable option depending on hardware support specifics for virtualization products like VMware, potentially necessitating the selection of a Fibre Channel SAN and thereby offsetting at least a portion of the anticipated cost savings. However, vendors are working very hard to improve virtualization-iSCSI interoperability, so pay attention to support matrices when investing.

2. Tape backup infrastructure

Tape drives have advanced to the point where their performance capabilities, based on traditional design concepts, are out of balance with servers and networks -- they're too fast! An oft-heard complaint is poorer backup performance following the introduction of new tape technology. The problem lies not with the tape drives but with the ability of systems to deliver data to the tape drives quickly enough to enable the tapes to stream. Reconfiguration of the backup environment, including introducing a disk cache ahead of the tape devices, is often the remedy.

Jim Damoulakis is chief technology officer of GlassHouse Technologies Inc., a leading provider of independent storage services. He can be reached at jimd@glasshouse.com.