As I explored last week, Amazon’s first foray into DaaS leaves a lot to be desired. Fortunately, you don’t have to look too far to find a service provider who knows how to get it right. This week, I’m looking at one of the first successful cloud desktop providers: dinCloud.
dinCloud started offering cloud desktops in mid-2010, a time when VDI technology was still very immature. Most organizations were struggling to scale even the smallest pilot projects. By getting on board early, dinCloud has built up some great domain expertise across the technical and service-related challenges that still cause so many cloud desktop and VDI projects to stumble. dinCloud is both a DaaS provider offering its service direct to end-user organizations out of its own data centers and a platform developer distributing its DaaS platform Ingram Micro, Tech Data, and Arrow along with 127+ VARs including CDW, Insight Enterprises, SHI, Tiger Direct, En Pointe Technologies, and others worldwide. Each is using dinCloud’s core technology behind its own cloud desktop service offerings.
There’s only so much you can do with DaaS unless you are able to move other data center services to the cloud in addition to cloud desktops. dinCloud provides core cloud services in the form of Windows Server (2008 R2 and 2012) and four flavors of Linux server, as well as D3, dinCloud’s file storage service, which supports Amazon’s S3 APIs. dinCloud currently has two US data centers, located in Los Angeles and Chicago. For customers looking to host desktops in other locations, dinCloud has partnered with the world’s largest colocation company, Equinix, which has 110+ locations worldwide and can support customers requiring 500 virtual desktops or more out of any of these sites in as little as 45 days.
dinCloud currently offers two enterprise-class DaaS platforms: dinHVD and webHVD. dinHVD, dinCloud’s first-generation DaaS platform, which has its roots in Dell vWorkspace, is scheduled to be phased out next year in favor of dinCloud’s internally developed webHVD. webHVD was created from the ground up as a purpose-built multi-tenant DaaS platform with its own internally developed remote display protocol. dinCloud also offers webHVD Home, a simplified version that omits Active Directory integration. In common with most DaaS platforms, to overcome the challenges of conforming to Microsoft’s draconian desktop OS licensing rules, dinCloud desktops are built on either Microsoft Windows Server 2008 R2 or Windows Server 2012 rather than Windows 7. For customers demanding the greatest flexibility and operating system compatibility, including 32-bit support, dinCloud can offer Windows 7/8. Microsoft’s licensing rules being what they are, though, dinCloud only offers Windows 7 in environments with a minimum of 150 seats, and at a significant premium over the cost of a Windows Server–based desktop.
Whereas Amazon WorkSpaces offers only two hardware specifications, dinCloud customers have far greater freedom of choice. They can create their own custom Windows Server–based virtual desktops from any combination of settings, from two to eight vCPUs, with 3, 4, 6, or 8 GB of memory and from 60 GB to 200 GB of storage. Windows 7 desktops don’t have quite the same flexibility: they are restricted to just two vCPUs and 3 GB of memory, although dinCloud says it can support requests for custom configurations should customers request them. One important benefit that dinCloud offers is that, as customers’ requirements change, dinCloud can readily adjust VM settings to add or remove vCPUs and memory through the dinCloud management console, with billing adjusting automatically.
dinCloud servers are two or four-socket machines using sixteen core AMD processors, 512 GB of memory, and either 2 x 10 Gb/s or 2 x 40 Gb/s network uplinks running KVM. dinCloud Windows Server 2008 R2 desktops start at $40 per month for two vCPUs, 3 GB of memory, and 60 gigs of storage. dinCloud has chosen not to offer a one-vCPU desktop, stating that two vCPUs deliver a better user experience and smoother display updates. In comparison, Amazon charges $35 per month for a single vCPU and 3.75 GB of memory. At the top end of the scale, dinCloud charges $398 per month for eight vCPUs, 8 GB of memory, and a full 200 GB of storage. dinCloud pricing is all-inclusive, with no additional charging for network bandwidth in or out, unlike Amazon, which charges for all non-display traffic leaving Amazon’s cloud. On paper, dinCloud’s pricing is somewhat lower than that charged by Amazon, although given the degree of variation between different service providers’ vCPU specifications, without benchmarking it is impossible to say which provider offers the best performance for a given specification. Unlike Amazon, which has no minimum purchase requirement, dinHVD has a ten-seat minimum. However, if this seems a bit much for a quick proof of concept, dinCloud, again unlike Amazon, provides a fourteen-day free trial service for potential customers.
If you need a real desktop OS, a Windows 7 desktop with two vCPUs, 3 GB of memory, and 80 gigs of storage would set you back $65 per month with a one-time setup fee of $50, but only for a minimum order of 150 seats. Small numbers are possible, but unlikely to be economical. Microsoft places highly restrictive rules on service providers’ governing the use of desktop licenses in VDI environments, which prevents the use of any shared infrastructure. Dedicating redundant physical hardware to what would otherwise be a straightforward cloud workload that could be cohosted on any server adds to any service provider’s costs significantly. If nothing else, dinCloud’s pricing for Windows 7–based DaaS instances and its 150-seat entry point shows just how hostile Microsoft’s VDI licensing rules are to service providers, stifling what could be a large and growing industry.
By default, all desktops and servers exist within a single dedicated virtual data center hosted in one of dinCloud’s two physical data centers, but there’s nothing to stop customers from creating multiple virtual data centers to separate production and nonproduction activities, or from hosting instances in multiple data centers in an active/active configuration to create a service offering higher availability than might be possible with a single data center implementation. However, using this configuration, some users might suffer from poor performance if they are connected to the more distant data center. Alternatively, dinCloud can replicate environments between data centers in real time. The one that dinCloud lacks, though, is an equivalent of Amazon’s Availability Zone service domain to provide the highest possible level of isolation as a single site. However, given dinCloud’s partnership with Equinix, it should be possible to identify multiple data centers close enough together to enable an active/active configuration without any performance concerns.
In common with many other DaaS vendors, dinCloud provides the option of licensing Microsoft Office as part of the core desktop platform. Office 2010 Standard is available for an uplift of $20 per month, or for $25 per month if Office 2010 Pro is available. Customers looking to do more than just run Office can import their own custom desktop and server virtual machine images and manage them through dinCloud’s web-based management console. Regardless of which option is taken, the core package includes initial on-boarding support as well as monitoring and alerting services, and 5 × 8 help desk support in the customer’s time zone, with the option for extended support available for a fee.
The standard dinCloud SLA offers 99.95% availability (about five minutes of downtime per week), with a 10% service credit for failure to meet this agreement, extended to a 25% service credit should availability fall below 99%. dinCloud calculates availability per month measured at the edge of dinCloud’s network, and measured and aggregated in five-minute intervals. I have seen many well-engineered distributed desktop environments that have failed to achieve this level of availability, mainly through change-induced incidents rather than any form of technology failure. With this in mind, it is worth reiterating that one of the benefits of VDI is the ease with which it is possible to roll back to previously known good desktop images in the event that a change to desktop configuration has untoward consequences. To this end, dinCloud automatically provides ten-day rolling backups of all machine images and data, above and beyond any archiving that the customer might undertake. Interestingly, dinCloud has chosen not to provide access to these backup snapshots from within the customer management portal. Moving these backups beyond customer control provides an additional level of protection, stopping hackers, or for that matter, rogue employees, from either deleting backups or overwriting current good data with an out-of-date copy. Very importantly, dinCloud has no access to the content of these backups: all data at rest and in motion is AES-256 encrypted.
All dinCloud management services are available through dinManage, a web-based control panel allowing the customer to monitor and manage services, report on utilization, and open support tickets. dinCloud is also actively developing its own management API, allowing end user organizations and service providers to integrate dinCloud with their own internally developed management platforms. Its current form, the dinManage API, supports quite comprehensive virtual machine management, including creating new virtual machines; managing system state; resizing and changing storage allocation; firewall, VPN, and network management; and limited data center reporting functions, as well as supporting VNC console access.
Although I had the option of setting up the service myself from the dinCloud management console, I chose to work more closely with dinCloud to understand the level of help available to new customers. dinCloud has approached the onboarding process with a hypothetical customer in mind: a midsize professional services organization with high expectations but limited in-house IT support. Taking this into consideration, it has hidden typically complex tasks behind simple web forms, enabling the customer to rapidly set up a VPN connection to the dinCloud data center and integrate with Active Directory with no real technical knowledge beyond naming and IP addressing information.
Automated setup assistance aside, dinCloud supports the customer through license management questions, creation of standard desktop images, and application installation. The support that dinCloud offers for license management and application installation is very welcome. Software license management in virtual environments, especially end user software, is still shaky ground for many organizations. dinCloud has several dedicated license management specialists on staff to help customers navigate this area. dinCloud also has substantial experience overcoming the occasional arcane challenges of getting desktop applications to work just right when running on a server OS, which can make all the difference between a smooth installation and late nights at the console.
dinCloud provided me with access to both dinHVD and webHVD. From an end user perspective, there’s very little to say about dinHVD. Based on Dell vWorkspace, dinHVD uses Quest Software, developed EOP remoting protocol, which delivered the expected “just as good as local” performance running out of dinCloud’s west coast data center, handling full-screen 1080p video without any obvious issues. From an end user perspective, dinHVD offers an experience an order of magnitude better than that delivered by Amazon WorkSpaces.
Looking to dinCloud’s future, I spent most of my time with its new webHVD platform. To support this, dinCloud has developed its own universal HTML 5–based client as an extension for Google’s Chrome browser. This is a good pragmatic starting point—Chrome is rapidly becoming the most widely used desktop browser and does not require desktop admin rights to install in Windows. While Chrome delivers excellent endpoint coverage, it does have its weaknesses. One of the biggest limitations of the Chrome extension is that it is only able to offer a single monitor support. This is something inherent to Chrome rather than anything of dinCloud’s making; the Citrix Receiver for Chrome has just the same weakness. To address this and other Chrome-related challenges, dinCloud is developing native clients for Windows and Linux. dinCloud has recently released a native Android client, and it has an iOS client for iPad in the works. dinCloud uses the open-source VP8 video codec to drive its remote display protocol; in contrast, Citrix uses H.264 in HDX 3D Pro. While dinCloud hasn’t been in the promoting business anything like as long as Citrix, this may turn out to be a better long-term choice. Both H.264 and the newer H.265 are owned by the ITU-T Video Coding Experts Group, a consortium of companies that collectively charge licensing royalties on every use of their video codec. VP8, on the other hand, was originally developed by On2 and was acquired by Google in 2009. Following some encouragement from the Free Software Foundation, Google released VP8 to the open-source community, adopted it as the standard video codec for YouTube, and subsequently rolled it into the Chrome browser. YouTube aside, VP8 has not seen the same widespread use as H.264. However, its successor, VP9, is tipped to become the most widely used 4K video compression codec in preference to H.265, which is considered by many to be too CPU-hungry for mobile adoption. The one downside of VP8 is that it is a lossy compression system. Going forward, however, VP9 does support lossless compression. With VP10 already in development, dinCloud has a clear long-term path for improving video performance already mapped.
On first starting webHVD, the virtual desktop defaulted to a screen resolution of 2560 × 1600, higher than the 1860 × 1080 than my desktop monitor supports. Rather than attempting to display the desktop at full resolution and leaving me needing to pan around the screen to access the Start menu, the dinCloud client shrank the display down to fit into the viewable area on the monitor. While this left everything looking small, it gave me a screen good enough to allow me to adjust the screen resolution down to something my monitor could support. Adjusting the screen resolution did reveal one small problem: whereas most desktops use a screen format ratio of 16:9, the webHVD desktop uses the less common 16:10 format, which leaves black bars on either side of the screen.
On first pass, the webHVD desktop felt more responsive than the Amazon WorkSpaces desktop. I didn’t attempt any formal benchmarking, but activities such as static webpage rendering appeared to be more immediate on the dinCloud desktop. Video performance suffered on webHVD compared to dinHVD. While a 1080p video was outstanding on dinHVD, Amazon WorkSpaces scraped by with a passable 240p. The same 240p video on webHVD struggled badly, stuttering and dropping frames regularly. I took the problem back to dinCloud, and they quickly diagnosed that my demo environment has been set up on a non-production server with some clunky old spinning disk storage. Fifteen minutes later, I had a newly provisioned desktop on a production server with SSD-based CEPH storage. Problem solved! I cannot say that its video performance was as good as dinHVD’s, but it more than surpassed that of Amazon WorkSpaces, happily running a 480p video without a single hiccough. The one remaining issue is the VP8-based lossy compression. While not a problem with video, compression artifacts are noticeable in static images. These were in no way severe enough to impact user experience when working with basic business graphics, such as PowerPoint creation, but they were unacceptable for Photoshop and completely out of the question for use in medical imaging. Even with these limitations, dinCloud is seeing considerable success with webHVD, reporting multiple deals of 1,000 seats and more.
If webHVD and VP8 were the only options available for dinCloud customers, I might be concerned as to its likely success. However, with a clear path forward to lossless compression with VP9 and beyond, and dinHVD with the Quest Software–sourced EOP protocol available as an interim solution, dinCloud looks to be on solid ground. It is already planning, in what will be a near-seamless migration, to move current dinHVD customers over to where the HVD will be when the time is right, ensuring that all customer data and desktops can be migrated intact. I look forward to revisiting webHVD as it continues to improve the platform.