|
TELECOMS AND SOFTWARE NEWS
|
|
|
Virtensys I/O solution: tactical versus strategic virtualization
Roy Illsley
Virtensys I/O solution: tactical versus strategic virtualization
Improvements in server virtualization are driving demand for improved network performance; I/O is now the performance bottleneck. This has forced many organizations to consider moving to 10G Ethernet. I/O virtualization represents one approach to resolving the problem. If combined with a unified communications (UC) strategy, this could help CIOs further reduce their management costs. I/O virtualization provides a tactical solution to increased flexibility and power savings
The typical I/O card and switch account for up to 30% of the power consumption of a standard server: improving their utilization represents a significant potential saving that server virtualization has not directly addressed. Another element of server virtualization that has created complications for organizations is the vexed question of how to ensure the data can be supplied to the virtual machine (VM) fast enough to avoid becoming a limiting factor in terms of performance. To highlight this limitation, the new Intel Xeon 5500 (Nehalem) processor is capable of handling 40Gbps, which means that 1G Ethernet or 1-2G Fibre Channel ports can not keep the processor fed with data. Intel and other chip manufacturers addressed this issue by making the VM able to operate with direct I/O, which raises the question of VM mobility: if a VM has a direct I/O access path to a physical I/O device, then moving that VM to another server involves breaking and making these I/O connections with every move.One approach is to virtualize the I/O , which can be achieved in a number of different ways. However, according to Virtensys (an I/O virtualization vendor), the net result of this approach is that it can reduce the number of physical ports required by up to 50%. In a fully loaded 42U rack, that can equate to a saving of approximately 2.1KW of power (although we believe this is an optimistic theoretical value). This saving is doubled when you also account for the reduction it has on any cooling solution: typically it takes 1W of cooling for every 1W of power consumed by the IT equipment. For many organizations, the need to increase network performance equates to increasing network capacity, which currently means upgrading the network. Such upgrades are expensive, particularly if your usage is variable: for example, if you use 1G for most of the day, and only require 6G at peak times, then moving to 10G Ethernet represents an expensive approach to solving the problem.I/O virtualization can help address these tactical issues as it can allow the fine-tuned allocation of bandwidth to those applications that need it, when they need it. Using this kind of multiplexing approach can reduce the under-utilization of resources that moving to 10G Ethernet would produce in the above example. A unified fabric approach to I/O management represents a more strategic solution.
A typical data center environment supports two to three parallel networks: one for data, one for storage, and possibly one for server clustering. This adds up to three times the headache for many CIOs, but a unified fabric approach consolidates these different types of traffic onto a single, general-purpose, high-performance, highly available network. However, one of the big questions about unified fabric is that, while it may greatly simplify the network infrastructure and reduce costs, in order to do this it must be intelligent enough to identify the different types of traffic and handle them appropriately.In addition to its potential to reduce TCO through simplification of the infrastructure and subsequent integration and automation of network, storage and VM operations, unified fabric supports broader data center virtualization by providing consistent, ubiquitous network and storage services to all connected devices. If the management issues of supporting these multiple networks are viewed from a strategic perspective, then moving to a single 10G Ethernet solution based on a UC approach represents a more long-term solution. But how interoperable are these solutions? Cisco's UCS, HP's Virtual Connect, and IBM's Open Fabric Manager are all examples. Yes, they have the ability to save power and energy by reducing the cabling, management overhead, and improved performance - but ultimately they lead to proprietary lock-in.
About:
This article is an extract taken from Ovum's Straight Talk service. This daily email bulletin provides our expert's views and opinions on important news and events in global IT and telecoms. If you have a comment or question regarding this article then please submit your details here:
If you would like to find out more about Straight Talk please contact StraightTalk@ovum.com
If you would like to find out more about Ovum services then please click
here for details
|
|
|
|
|
|
|
|
Search
|
|
|
|
|
Contact Us
|
|
|
Expertise
|
|
|
|
|
