Grid Engine Gets a PaaS
This morning Univa announced a partnership to allow Grid Engine users the capability to weave in Eucalyptus on-premise cloud management software. Univa, which acquired Grid Engine in January, claims that this ability will allow users to “exploit the benefits of dynamic, scalable and self-serve cloud systems within the backbone of their production compute and data analysis infrastructure.
According to Univa, this could open the door to number of benefits for HPC users, including the possibility to extend cloud projects into the backbone of large data analysis infrastructures, the ability to work with self-serve provisioning tied to workload requirements, opportunities to burst into Amazon’s IaaS for extra resources, and the capability to manage multiple sites (on-premise and cloud systems) with ease.
As Univa CEO, Gary Tyreman stated, this partnership is “all about enabling a Grid Engine PaaS (Platform as a Service) as Grid Engine’s primary use is in technical computing.”
To better understand how Grid Engine users might be able to take advantage of the partnership or simply get on board with the practice of “cloud bursting” for additional resources during peak demand, we asked Tyreman about what management complexities are involved before and after–and what this means for those HPC users still “on the fence” when it comes to cloud computing.
HPCc: What is the process for users to integrate Eucalyptus into their existing Grid Engine environments? What is involved, how long does it take, what type of assistance is given?
Tyreman: The process is straightforward and is similar to what we done with Amazon AWS. In fact, we talk to Eucalyptus in the same manner as Amazon EC2. The integration process, whether adding Grid Engine to an existing Eucalyptus environment, or alternately, adding Eucalyptus to an existing Grid Engine environment is the same.
We make the assumption that Eucalyptus has been installed and configured, so in our view the IaaS is ready. Today, we do nothing to simplify this process, although this is a potential area of collaboration in the future. Once Eucalyptus is configured the next step is to install and configure UniCloud, a one-time process that takes fewer than 60 minutes. Provisioning and configuring Grid Engine at this point is readily done via UniCloud’s API, GUI or command line. In effect, this is the point at which one pushes out the Grid Engine “bits” and configures the VM. Eucalyptus makes the decision where to place the “machine”.
Essentially what we have done is created a Grid Engine PaaS that runs on top of any cloud platform – public or private – and by using automation we have cut the time required by 99% (from about 4 days to 1 hour).
With respect to services, Univa can assist with this setup, but we’re confident that the automation creates a level of simplicity such that outside help won’t be required.
HPCc: Will you look beyond Eucalyptus to other cloud software packages, including open source ware like OpenNebula or others?
Tyreman: Yes. In fact, the astute will have noticed that RightScale will join us in June at the upcoming Grid Engine User Summit. This is all part of what we call “One Click HPC”. We have a number of pending announcements in this regard. Today we are focusing on the players we believe offer Grid Engine users the greatest synergies.
HPCc: Related to the above question—why Eucalyptus, what are the details of this partnership?
Tyreman: Eucalyptus was among the first partners in this program for Univa for three simple reasons. First, they enjoy tremendous market presence with over 25,000 existing clouds and several thousand downloads each month. Combine that with over 10,000 data centers using Grid Engine and one can quickly find customer overlap. Second, Eucalyptus provides a graceful integration point between our strategic partners. Third, there are simply too many moving pieces for any one company to overcome, and as history teaches us, best-of-breed wins most often. For Univa, this partnership will allow us to focus on driving more value from an existing Grid Engine investment, while at the same time this will permit Univa to push the innovation curve.
HPCc: You have stated that your partnership will help reduce complexity although the terms were very general—what problems are users facing and what specifically will this capability provide?
Tyreman: The bottom line here is complexity, familiarity and automation. The average IT admin in technical computing has had little to no experience with virtualization, the building block of clouds. The installation, management processes and impact of the technology are not well understood – in a practical way. Therefore, integrating these technologies and platforms can be trial and error.
Everything Univa has done with UniCloud could be completed manually and could take up to 4 days. Cutting that time to minutes doesn’t just buy time (which is certainly of value) it buys know-how and capability. The value clouds bring to this market are documented and real. Being able to capture it, we felt, remained a challenge.
Our goal was to bring best-of-breed products together and integrate them like one would normally see from a suite. This puts the capability into the hands of the many.
HPCc: How are users going to be charged for this additional capability?
Tyreman: Univa Grid Engine users will have access to this capability. Open source users need to add UniCloud which is priced per managed core. Eucalyptus costs would be on top of this.
HPCc: Do you expect that this offering will tempt more users to consider the possibilities of cloud computing for their current environments—that is, if they see it is available, will it drive them?
Tyreman: All technical computing users face some very serious and real issues today that are only going to get worse in the future. Simply put, there is too much growth in the computational and data analysis needs of these companies and little to no growth in the budgets or performance increases of core building blocks. For example, the leaps in throughput that were driven by clock speed improvements are gone, and in its place complexity has exploded with increasing core counts.
The struggle to get more done in the same time will become greater each year. Therefore, wringing every ounce of performance out of the system will take on new meaning as long as time remains the key variable. This is where cloud comes in.
Cloud promises to offset the pain in the future by enabling companies to execute new compute strategies.
Our view is that this capability will facilitate Grid Engine users to begin to execute such strategies sooner and more completely than competitive products.