Aprius IOV Technology Evaluation Platform
IT professionals should be starting to look at I/O Virtualization (IOV) as a way to better utilize data center resources to further drive out costs and increase flexibility. IOV essentially establishes an I/O gateway that allows multiple high bandwidth interface cards to be consolidated inside of it and then shared to a group of servers that connect into the gateway. Sunnyvale, CA based Aprius last month announced the availability of their first entry into this space.
Friday, May 28, 2010
Aprius has taken the unusual approach of more accurately setting user expectations in the product name itself, calling the product the “IOV Technology Evaluation Platform”. It is designed to be tested by OEMs and select end-users as they are developing IOV aware technologies and strategies. One of the developments required before IOV can reach the point of wide adoption is for interface manufacturers to add Single Root I/O Virtualization (SR-IOV) to their cards.
George Crump, Senior Analyst
Aprius is a client of Storage Switzerland
The SR-IOV specification is a standard for a type of PCI pass-through which natively shares a single device with multiple guests. SR-IOV allows a single device to be presented as multiple devices to connecting servers, complete with their own configuration information. SR-IOV adoption is being driven by more than just the IOV vendors; it’s also needed to better allocate resources to hypervisors that control virtualized server environments. While only a handful of interface cards today have SR-IOV capabilities it is expected that within the year many more interface manufacturers will have added that functionality.
What makes the Aprius platform unique is their selection of Ethernet as the connection method between the servers and the IOV gateway. As we discuss in our article “Comparing I/O Virtualization Technologies” there are two common connection techniques; PCIe and Infiniband. While both have merit, the technique chosen by Aprius may be the most interesting. Leveraging a capability that they call “PCIe over Ethernet” it can tap into an infrastructure that is already widely deployed and well understood. Today, these connections require a specialized 10Gb Ethernet card in each server, which is cabled directly to the top of rack into the IOV gateway as pictured below.
As the platform and the environment develop, the advantages of PCIe over Ethernet can be further exploited. The first use case is leveraging the existing Ethernet switch infrastructure to facilitate connections to the IOV gateway. This could possibly move the gateway from a top-of-rack device to an end-of-row device.
One of the more interesting use cases though, may be in the deployment of the technology in blade servers. In this application, the Aprius PCIe over Ethernet initiator can be embedded into the blade system itself and then connected to the IOV gateway through the blades system’s 10Gb Ethernet connection. This could be a welcome addition to space- constrained systems that are forced to use a finite number of proprietary interface cards.
The current evaluation platform is being demonstrated in three basic configurations. The first is utilizing an Exar-Neterion 10GbE NIC with SR-IOV support. This configuration allows the 10Gb Ethernet bandwidth to be shared across all of the connecting servers. The gateway then becomes a rack I/O expander. Essentially any card supporting SR-IOV could be shared simultaneously with the connecting servers. Non-SR-IOV cards would have to be shared one at a time or one port at a time. For example, Aprius has demonstrated allocating each port on a quad-port Q-Logic fibre channel card to separate servers connected via PCIe over Ethernet.
The second configuration for evaluation is the flash SSD appliance. Leveraging a single SR-IOV capable LSI 6Gb SAS controller in the gateway, all the connecting servers have shared access to SSDs in an attached disk shelf. It’s important to remember that SAS has four channels per controller so the total bandwidth from the SAS controller to the SSD is 24Gb per second. There would also need to be storage virtualization logic in the attached disk shelf to provide shared access to multiple flash drives. This configuration allows for the high end bandwidth of SAS to be fully realized and the expense of SSD technology to be shared across multiple connecting servers.
The final configuration for evaluation builds on the flash SSD appliance and leverages SR-IOV capable storage network adapters like FCoE, 40GbE (2011) or 8Gb FC to connect to a back end storage system. Two of the PCIe slots in the gateway will be populated with these storage cards and connected to the Enterprise SAN. The servers will still connect via the PCIe over Ethernet connection but now have access to the SAN as well as the SSDs. In this configuration the file systems on those servers could use the flash-presented storage as large read/write caches for high speed delivery of very active data, which could be flushed down to mechanical storage later.
While not in an IOV environment, Storage Switzerland has tested the use of proportionately small amounts of SSD as read/write caches and found that they can provide significant improvements to overall performance. The challenge has been that only 100GB or so of this cache is typically needed, wasting very expensive SSD capacity. IOV potentially provides a way to better share this previously wasted capacity.
The Aprius IOV Technology Evaluation Platform provides OEMs and end user customers with a viable way to take a first look at IOV, one we believe is an important step to take. IOV has the potential to provide a tremendous advantage to OEMs that support it and to companies that leverage it. However, since it is an infrastructure product, it must be carefully integrated into the plans of the enterprise. Aprius, by delivering the solution initially in this manner, provides OEMs and customers that capability.
Briefing Report
