The most significant part of VMware PEX, for me this year, was the Solutions Exchange floor and the rather small number of vendors. My focus was on convergence of compute and storage resources. This appears to be a popular path. There were a few things along the way, other than cheap swag, that caught my eye. One interesting conversation involved FusionIO. They validated that many customers concentrate more on storage space instead of performance and that this is not good. Some more progressive enterprises are very focused on performance. For instance, eBay actually measures costs based on url per kilowatt.
LeftHand introduced the idea of virtual storage appliances in early 2008 and has been improving ever since. A former coworker of mine has actually been speaking about the idea of convergence for quite some time now. The introduction of the VMware Virtual SAN seems to validate the idea of convergence.
Listed below are the significant finds in no particular order:
The vSTAC VDI P-Cubed is a starter appliance for Horizon View deployments It is all-inclusive in licenses with the exception of the Windows Desktop licenses. It is only meant for View. All of their other products appear to be purpose built for video surveillance. The appliances are built on Dell R720 hardware and it appears as if the hardware is supported directly by Dell, which may cause confusion in when calling support.
The OmniCube is based on “a 2U standardized x86 server from a tier 1 server manufacturer.” It appears that Simplivity supports the hardware directly and will ship spare parts as needed. It currently supports vSphere only and is tightly integrated into the vSphere (fat) Client. They are currently working to get it compatible with the Web Client. The differentiator is that incoming data is processed in real-time by the OmniCube Accelerator
HP StoreVirtual Virtual Storage Appliance
The granddaddy of VSAs, Scott Lowe wrote on Techrepublic about the LeftHand VSA in 2008. The current version, now called the StoreVirtual VSA is “free” with the purchase of select HP Gen8 servers. It currently supports both vSphere and Hyper-V and includes automatic data tiering, which is different than the flash acceleration that is leveraged in the other products. One unique differentiator is that it can be coupled with hardware based StoreVirtual 4000 arrays to create different tiers of data storage.
VMware Virtual SAN is expected to be GA March 6th, 2014. As expected, it supports vSphere only and is integrated with vSphere Web Client. It is already a part of the ESXi 5.5 hypervisor and is activated with a license key. It is a “flash optimized” solution, like Simplivity, Pivot3 and Nutanix. It offers policy based management at the object level in vCenter. It also gives you a choice of hardware, as long as the disk controller is on the HCL.
Although not actually present at PEX, Nutanix was in San Francisco last week, so I am including them here because I think they are a leader in the converged space. It currently supports vSphere and Hyper-V. Based on Supermicro hardware, it offers multi-node blocks in a 2U space. One of the key differentiators is that it uses Map Reduce to manage data and localize it to the node that hosts a VM. It uses VMDirectPathIO / SR-IOV for connection to the disk controller.
FusiionIO is flash based cache and performance acceleration. There are software and hardware solutions to deliver lower latency and lower power costs. The ioControl array uses PCIe Flash instead of SSD for lower latency. One interesting thing is that they mentioned per kilowatts are being measured by eBay. The message is that performance and latency are more important than space when it comes to storage.
Although not advertised, Supermicro is the hardware used by Nutanix and FusionIO. The Twin family of servers offer two to four nodes in a compact 2U space. Theoretically, you could build your own using Supermicro hardware and VMware vSAN.
A relative newcomer to the US, Bull appears to have a full line of servers. The eye catcher is their scale-up Bullion Server which can scale up to 16 sockets by stacking four quad-socket boxes. It has lots of RAM (see photo below) and plenty of room for local DAS, networking and IO expansion. Because the modules are stacked, losing a module does not result in a VMware HA event. There is also a VSPEX Reference Architecture that includes Bull, VNX storage and Brocade Networking. (http://www.bull.com/storage/emc-vspex-powered-by-bull.html)
I really don’t do much VDI. In fact, most of my conversations with customers are more about mobility. Many times, traditional VDI is not a good fit for the use cases. All of the marketing fluff about ROI/TCO takes into account “soft costs” that many do not accept in this economy with already-thin staffing. However, the conversation often turns to the end-goal and use cases. Many times, something like XenApp is a better fit instead of View or XenDesktop.
That being said, the soft cost savings MAY be realized in engineering and architecture environments that use higher-end CAD. Localizing the desktops to the file shares drastically reduces the file operation times. The task of opening or saving a huge CAD file could take significant amounts of time. Now we are talking about the labor costs of design engineers or architects. The issue is that they need to have a better GPU than what comes in a typical server and, until recently, there was no way to share a GPU. Offering up desktops to these types of uses often meant a 1:1 ratio, using blade PCs or workstations placed in the datacenter. Enter things like HP’s Accelerated Graphics (http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA4-1701ENW&cc=us&lc=en). This allows for multiple GPUs in a single host for pass-through 1:1 GPU configurations, hypervisor-level “Software Virtualized GPU” and Hardware Virtualized GPU configurations. Although it is pretty expensive, it allows for an acceptable user experience when rendering 3D and 4D.