Network Computing is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Astute Networks - Attacking Server Virtualization I/O Bottlenecks

For cash-strapped IT organizations, server virtualization seems to be a Monopoly-like “get out of jail free card” with many compelling benefits, such as better use of IT assets and energy savings. But at some point, as the number of physical servers drops and the percentage of virtualized workloads and applications rises, a perhaps strange and unexpected thing can happen: application service-level performance suddenly goes to hell in a hand basket. I/O bottlenecks are the problem here, and Astute Networks has a solution in the form of a flash-memory-based appliance.

The problem that Astute Networks, along with a number of other companies, attacks is a serious one. Often in IT, when we make the decision to adopt a new technology
(such as server virtualization or cloud computing), unexpected obstacles occur midway through the process, hindering progress. The challenge is to solve those problems so continued adoption of a valuable technology goes on.

The problem of virtualization-related I/O bottlenecks is conceptually easy to explain. In non-virtualized IT environments, one application tended to run on one physical server and there was usually more than enough I/O bandwidth available to meet the response
time requirements of applications. However, in the most ideal circumstances, you want to put as many virtual machines on a physical host server as possible. As the number of virtual machines rises on a single physical system, the aggregate I/O requirements of related applications (and that includes not only production applications, but also backup applications) can reach a point where a bottleneck occurs that seriously degrades
application response time performance.

That tipping point may not be easy to determine a priori; one day you are doing fine, while the next day performance is terrible. This is a classic “knee of the curve” problem where the point beyond is one in which response time decays dramatically. However, understanding that you have an I/O queuing curve problem is a lot easier to detect than to solve. The simplest might be to backtrack and undo some of the virtual machines that you have already deployed, but this is unpalatable in the long run as it limits users to
only some of the benefits of server virtualization that they had hoped to capture.

Astute Networks’ answer to I/O bottlenecks is a software/hardware appliance called the ViSX G3. The appliance’s hardware is based on flash memory modules and uses what the company calls its DataPump Engine; a multi-protocol network-attached storage offload processor whose hardware, and software drive the flash memory module. The DataPump Engine processor accelerates TCP/IP network traffic, as well as iSCSI
virtualized data center traffic, while at the same time using flash memory to give a sustained high level of IOPS performance.

Competing flash solutions obviously exist, but Astute Networks feels that locating the appliance in the network as a complement to existing storage is superior to locating the flash module on the host or as DAS or network storage, as do some other vendors. It
feels as if using internal PCI flash on host storage has limitations, such as fixed level of performance and performance limited to only one server. External flash as DAS storage has one of the same limitations as all flash is dedicated to one server. The other flash memory alternative, putting flash as a tier of SAN on network storage, does not get the most out of flash memory. I am sure that its competitors would disagree with this
assessment. The market will decide who is right, but Astute Networks makes a strong case for its approach.

  • 1