InsightaaS perspective: 451 Research is one of the world’s leading sources of insight into cutting edge technologies (including areas that are important to InsightaaS and our principals, such as cloud, collaboration, analytics, and sustainable IT). The company is also a leading supplier of information on mergers and acquisitions in the IT industry, which it publishes in its subscriber-only M&A KnowledgeBase; pieces designed for public consumption can be found on 451 Research’s Inorganic Growth blog.
As regular visitors to this site are aware, InsightaaS.com works with 451 Research to bring occasional thought leadership pieces to our readers. In this piece, we are privileged to present insight from 451 founder and distinguished analyst John Abbott on a critical issue in IT management strategy — the use of and outlook for Desktop as a Service, or DaaS. In today's multi-device, mobile world, DaaS is touted as a means of applying IT disciplines to an increasingly unruly endpoint device population. To date, DaaS has had limited market acceptance — but as Abbott points out here, both the range of supported applications and the ecosystem of DaaS providers is expanding, and we may be approaching the point of "mainstream acceptance of DaaS for the first time." If this is the case, IT management will need to examine the potential to expand cloud and infrastructure focus to extend security and data management to all connected nodes, inside and beyond corporate walls.
Note: if you are interested in obtaining a subscription to the 451 Research Infrastructure Management program, please contact 451 directly, or contact InsightaaS at firstname.lastname@example.org.
VMware has always insisted that it wants to virtualize any application, even the most demanding enterprise and mission-critical ones. This week it made progress toward supporting another application sector that was previously a no-go area for virtualization — 3-D graphics. At NVIDIA's GPU Technology Conference (GTC) in San Jose, California, the two companies announced a joint development agreement that will see NVIDIA's GRID vGPU technology integrated into VMware ESX, enabling up to eight virtual desktop users to share a single graphics chip. And although this effort won't be completed until toward the end of this year (with general availability not until 2015), VMware did make its current vGSA and vDGA implementations of GRID available on the VMware Horizon DaaS (desktop as a service) Platform, which will enable 3-D graphics on virtualized desktops and applications to be delivered through the cloud — at least as soon as the DaaS providers that use the platform upgrade their back-end systems. NaviSite will be the first to do so, and has plans to make it available 'later this year.'
NVIDIA rolled out its first take on GPU virtualization at the GTC conference two years ago, but only after a five-year development effort. Graphics chips, with their much higher core and thread counts, are more complex to virtualize than CPUs. The first integration efforts were with the Citrix XenServer hypervisor. Citrix remained ahead of the curve (presumably because it was more amenable to adding the necessary new code to the hypervisor) and in December 2013 announced full hardware vGPU sharing for XenDesktop 7.1 HDX 3D Pro with XenServer 6.2 SP1, in conjunction with NVIDIA GRID K1 and K2 graphics cards. VMware currently offers two interim graphics models while it works on full vGPU sharing. They are vSGA (virtual shared graphics acceleration) and vDGA (virtual direct graphics acceleration). The vSGA, available for Horizon View version 5.2 since March 2013, does provide shared virtual access to a GPU, but requires a VMware driver within each guest — with serious consequences for compatibility. The vDGA, introduced with Horizon View 5.3 seven months later, uses direct GPU pass-through to the guest VM via PCIe, which means only the native NVIDIA driver is required (and thus full compatibility with graphics applications). Of course, this means no GPU sharing — it's a one-to-one relationship and so an expensive option that doesn't scale well. Its intended market is for high-end graphics applications that require remote access to a dedicated GPU.
The VMware Horizon DaaS Platform is based on the technology platform VMware acquired from Desktone in October 2013 for $105m, supporting multi-tenancy, self-service provisioning and management for service providers looking to offer their customer DaaS. Desktone's service provider customers included Dell, Dimension Data, Fujitsu, IBM, NaviSite and Time Warner Cable at the time of the acquisition. VMware also decided to provide its own DaaS offering using the Desktone platform under the name vCloud Hybrid Service, or vCHS. That service went live on March 10, priced at $35 per user for a basic desktop package. The Horizon DaaS platform will now support for graphics-rich desktops and applications through NVIDIA's GRID technology, and potentially attract a broader range of customers as a result, from general-purpose graphics use through to engineering firms, automotive and retail product designers. The most likely scenario is that these types of users will take advantage of DaaS when project teams are collaborating or when contractors are working at remote locations. Graphics-intensive 3-D applications can be accessed from any location and from any connected device served. But GPU virtualization will also enable users to share graphics hardware resources at lower costs. Example applications already supporting GRID include Adobe Photoshop, Autodesk SolidWorks, PTC and TeamCenter — but it's also applicable to rich Web applications supporting HTML5.
NaviSite will be the first service provider to implement the technology and will extend its existing DaaS service to support graphical applications some time later this year. VMware itself and other DaaS service providers will follow. Initially, however, only the vSGA (shared) and vDGA (direct) graphics models detailed above will be available, each of which have their restrictions. Once the joint development on vGPU is completed, toward the end of this year (for early access users) but for general availability in 2015, full GPU virtualization will also become available to Horizon DaaS service provider users. That will allow provisioning up to eight users per GPU, with the GPU acceleration shared across the multiple virtual desktop — and (unlike the shared vSGA model today), all using native NVIDIA graphics drivers, ensuring full compatibility for applications. With vGPU, graphics commands go directly to the GPU, which then assigns the optimum size of memory to meet the requirements of each user. Critically, the upgrades from existing graphics models will be seamless, so DaaS users starting off using vSGA and vDGA won't have to make any additional investments.
To support vGPU service providers need to build out their own GPU resources inside their datacenter servers. They do this by installing server boards using NVIDIA's Kepler processor family. There are two choices: the K1 and K2. The K1 has four GPUs and 16GB of memory and supports up to 32 users — if the reseller price for the board is about $2,600 then the price per user comes out at roughly $60. (The user limit is designed to provide a minimum frame buffer size compatible with most applications). The K2 has two more powerful Kepler GPUs (with more cores) and 8GB of memory — it's designed to support up to 16 power users and designers when density is not the main requirement. The additional cores mean larger models and datasets can be supported. Both boards have an on-board streaming engine that offloads the CPU from encoding, which improves power efficiency.
NVIDIA is busy expanding its activities beyond the gaming and specialist professional graphics markets where it made its name. The catalysts for this were the invention of general-purpose computing on graphics processor units (GPGPU computing), through which its graphics chips were stacked into servers and applied to general purpose compute tasks, and the establishment of a programmable platform, CUDA, for developers. In the last few years, three new areas in particular have become established and are now growing well: big data analytics, cloud and computer vision. While applicable to professional graphics users, vGPU also looks likely to become a sales driver for NVIDIA's cloud efforts. A number of cloud providers — including Amazon, Nimbix, PEER1 Hosting, Penguin Computing and IBM's SoftLayer — now offer NVIDIA-based GPU computing as a cloud service. It won't be much of a leap for those service providers to add support for hosted graphical desktops.
Meanwhile, Amazon (which uses a variant of Xen as its hypervisor) introduced its WorkSpaces DaaS offering in 2013, and it has just become generally available. It seems likely that Amazon will make the connection with its GPU services at some stage in the future — although as it uses the Windows Server operating system for desktops, it won't have the flexibility of providers such as NaviSite that can offer a 'bring your own license' option for Windows 7 and 8 desktop users.
All this will of course boost server add-on sales for NVIDIA as the service providers build up their back-end infrastructure to support it. Traditionally servers didn't include graphics chips; now there's a reason to include them or add them. Cisco, Dell, HP, IBM, Lenovo and SuperMicro already offer the Kepler-based plug-in cards needed to support GRID either separately or pre-installed inside their servers.
NVIDIA now also sells its own GRID-compliant GPU server, the VCA Visual Computing Appliance, and has even developed its own graphics remoting technology for VCE, including the proprietary GRID stream protocol. This is also used in for NVIDIA's GRID Cloud Gaming platform for online gaming service providers, enabling high-quality games to be run on mobile devices. Both these efforts are relevant to the ongoing joint development work NVIDIA is involved in with Citrix, Microsoft and VMware.
NVIDIA stands as the neutral party in the battle to win market share and support a broader share of applications in both on-premises virtual desktop and hosted DaaS. VMware, Citrix, Microsoft and Amazon provide the infrastructure and (aside from Citrix) also offer direct cloud services themselves. It also supplies multiple competing hardware vendors, including Cisco, Dell, HP, IBM, Lenovo and SuperMicro, with boards to embed in their server products. NVIDIA's own VCA server is not really competing with these vendors, it's really a showcase and reference design for new technologies. NVIDIA's primary GPU competitor is Advanced Micro Devices, which acquired NVIDIA's direct independent rival ATI for $5.4bn back in 2006 — and since that acquisition AMD has perhaps been more interested in closer integration between the CPU and GPU than in developing the GPGPU as a potential rival to the CPU. AMD does support remote graphics and VDI for its FirePro chip range, but its efforts in these areas so far have remained low key.
|NVIDIA has been busy expanding the use cases of GPUs for several years and has built up an ecosystem of vendors that are now ready to support its move into virtual desktops and DaaS.||There's some market confusion over the many different graphics models supported by GRID and VMware, ranging from software emulation through to direct pass through. This should be simplified as the technology matures.|
|GRID opens up the server sector for NVIDIA, an opportunity that didn't exist at all a few years ago. Future server-oriented chips (Project Denver) may expand this further.||NVIDIA faces little direct competition because AMD's activities are focused in different areas — but it has to make sure it doesn't compete against its OEM partners with in-house technology developed in-house that overlaps with theirs.|