Australian IT systems integrator Frontline Systems announced Steam Engine, which is a cloud-based infrastructure provider with an eye on the HPC market. According to the company’s release, Steam Engine will offer “high performance computing infrastructure with zero capital expenditure and cost-effective short-term leasing options at a minimum of one month.”
The company claims that they are seeking to fulfill demand for this kind of infrastructure in key market segments, including “visual effects, geo-science, mining, biomedical and financial industries for bursts of rendering, simulation and process-intensive applications.” This focus is due in part to the fact that they see these areas as being best suited for what they offer since many of these segments are characterized by wide variations in demand and oftentimes there might not be the financial support for in-house solutions. They cite a number of animation studios currently using their service, although many are based exclusively in Australia.
When asked what their competitive advantages were and what made them relevant or different in the space, we did receive an answer that comes from others in the same arena, including mentions about scalability, expertise, relationships with vendors and general statements about not being the same. The company also noted “geographic proximity to customers when considering HPC payloads” as a point of difference between the others who vie for the same customers. While the last point they make is an interesting, albeit slightly vague, issue of contention, as a newcomer in the HPC cloud space, it is necessary to evaluate them in the context of the others that have come before—and the onslaught that will come once the HPC cloud matures.
To dig a little deeper into the company’s HPC cloud offering, HPC in the Cloud communicated via email with Stefan Gillard, the company’s commercial director and Michael Chanter, CTO for further insight.
HPCc: Although you hinted at some of the possible reasons, including demand among those with spikes computing need, why are you emphasizing the niche markets? What drove the choice?
Gillard/Chanter: Mainly because the markets are ‘HPC ready’, but also because the Frontline and Steam Engine teams have established track records in these industries at a deep level. Although the software workloads are different, they have similar processing requirements (ability to parallel, need for scale and spiky in terms of demand), and are market spaces where HPC is an accepted and understood paradigm.
HPCc: Most use cases or commentary points about Steam Engine are Australian-are you focused on customers in your backyard or beyond? What is your current versus projected reach?
Gillard/Chanter : We estimate that we will have 4000 nodes (48,000 cores) deployed in Australia within the next 12 months in Engine1 as well as appox 2PB of Tier 1 and Tier 2 storage as well.
We are currently working to deploy Engine2 into Vancouver B.C. comprising 3000 nodes (36,000 cores) by Xmas 2010.
We are currently finalizing plans for a deployment of over 4000 nodes (48,000 cores) into Mumbai India as well with Engine3. This will include approximately 2PB of Tier 1 and Tier 2 storage as well.
HPCc: What are the node sizes/architecture, x86-based?
Gillard/Chanter: Current architecture for nodes is entirely based on HP server platform.
Blade servers are
HP ProLiant BL2x220c G6 L5640 24GB (2P)
Intel® Xeon® L5640 (6 core, 2.40 GHz, 12MB L3, 60W) Dual Hex Cores
24GB Ram per Node
1GbE NC362i 2 Ports
2RU units are
HP Proliant SL2x170z G6
Intel Xeon X5670 (6 core, 2.93 Ghz, 12MB L3,90W) Dual Hex Cores
48GB Ram per Node
2TB of local disk per node
HPCc: What is operating system/s – Redhat, Microsoft, Debian, SUSE or all of them?
Gillard/Chanter All of the above as well as any flavor our clients may have tuned themselves.
HPCc: What virtualization software are you using?
Gillard/Chanter: OS and virtualization are customer dependent. The key architectures we support from a virtualization perspective are CitrixXen, Hyper V, VMWare.
HPCc: How are the nodes interconnected – Ethernet 1Gig or 10Gig or is it Infiniband?
Gillard/Chanter: At this stage Steam Engine is offering 10gig interconnections combined with HPC-specific switching architecture utilizing Arista and Brocade switching technology. The performance of a HPC solution is not just down to interconnect technology. HPC is a solution stack, and every consideration made around our networking layer, storage, interface, memory profile, application stacks influence total throughput of the environment. Our expertise in tuning all of these elements together effectively, provide us with a competitive value proposition when compared to other providers.
In terms of switching we are utilizing products from both Brocade MLX series, and Arista 7500 series product lines.
HPCc: Since you’ve indentified biomedical HPC users as a target market, the big question is what security provisions have been put into place to ensure financial data biomedical data is not ‘leaked”‘?
Gillard/Chanter: We have segmentation based on physical chassis, storage, network and virtualization segmentation. Segmentation can be physical layer to virtual layer depending on customer requirements. Steam Engine 1 infrastructure is hosted in Tier 3 Data Centre, which is one of the 5 data centers that are Federal Government certified. Our HPC environment is located in the same DC as the ATO.
HPCc: What is your disaster recovery plan and can you guarantee no data loss? if you have guarantees what is the penalty if system fails and data is lost?
Gillard/Chanter: DR is based on customer requirements. We have 2 different customer scenarios.
1. Digital Asset Management & Data Storage, and the other is
2. HPC compute where often the customer data is transient on the environment.
In Scenario 1 we have DC to DC redundancy, and disc to disc and disc to tape redundancy dependent upon customer requirements. In scenario 2, DR is not a primary consideration, as typically, the output is shipped back to the customer on completion.
HPCc: While Steam Engine is a new launch, Frontline is not—can you explain this relationship and state how long have you been in business?
Gillard/Chanter: As a new Frontline venture in the HPC cloud space, Steam Engine has been in business since March 2010, however as a Enterprise Infrastructure System Integrator we have been servicing Enterprise Infrastructure customers for over 19 years.
HPCc: What is the software stack, or do you bring your own?
Gillard/Chanter: We can either provide or the customer brings their own. All our services come with professional technical support for as long as you use the service. We work closely with you to ensure your application works as you would expect it to and guide you through the use of our services. Whether you are a first-time Linux user or an expert, we want to provide you with hassle-free cluster computing services.
HPCc: Why and how are you different from Amazon’s EC3?
Gillard/Chanter:
1. Higher specification environment ie you are accessing physical infrastructure not virtualized cluster instances.
2. More competitive pricing
3. Support and Assistance
4. Certainty of location and Data Location
5. Geographic proximity to customers.
6. Engineering support for localized customer requirements, as opposed to faceless cloud.
7. Our Infrastructure is specifically tuned to the workloads we offer our customer base. (i.e., our use of BlueArc storage tier 1 for the vfx and animation idustry which is known to be the benchmark for HPC storage requirements, compared to the storage IO specification of other players in the market including Amazon, is second to none).
* Note * We will continue to follow the company’s developments to see if they expand outside of their borders, both market and geographically-wise—and how they will manage to differentiate themselves in practice over a diverse set of use cases.