Scott Farrand, VP, ISS Platform Software, HP, talked to Prashant L Rao about HP’s Gen8 servers and the vendor’s focus on manageability, SSDs and green computing as part of its Project Voyager launch
There are a lot of innovations around manageability in Gen8. What led to your focusing so strongly on that aspect?
The engagement that we have had with our customers has driven us to this feature set. It has taken roughly a two year time window and we have spent a lot of time with about a hundred thousand customers and rolled up all of their concerns and categorized them. From this we came to the need for providing a lifecycle like approach for the conveniences, to get cost out, to help them out in getting the firmware updated and those kinds of things.
We have had embedded management solutions built into ProLiant servers for years. It goes by the name of iLO, which is an embedded processor. However, there’s a radical improvement in Gen8. The step function improvement is about the vastness of the monitored parameters that make it more serviceable and manageable.
The Gen8 servers are said to be optimized for SSDs. What are the implications of that?
The solution that we are providing is DAS. SSD drives are not a new notion here, their prices are ever decreasing and therefore they are in higher demand for their faster IOPS and faster stores. With this phenomenon happening in the industry, we have designed the controllers that manipulate those SSDs in Gen8 to be conducive and optimized for that. We had to change the algorithms for the improved speed as the traditional spindles were slower. We had to ensure that the data reliability was there. So the cache write through mechanisms are now flash backed up.
What kind of attach rate do you expect for SSDs?
We are anticipating modest uptakes on the order of 10% in the beginning. In ensuing years it will go up and get closer to 30-40% in a few years. It’s all driven by the application throughput.
Because of the Cloud, we are seeing an increased amount of use models of DAS vs. shared storage. Large unstructured data and Hadoop like phenomena are using DAS. Then there are internal applications for which organizations are are using x86 devices to do the storage rather than specialized devices.
Are there a lot of green aspects to this launch?
We have done a lot of work in this area, not only in this but also in previous generations. In this generation, one innovation is with the power supply efficiency, the second one is discoverability of power so that it is configured in a manner that doesn’t lose the redundancy that’s associated with it. Every aspect of the design is power efficient and controllable by the management co-processor. E.g. on the blade-based solutions there’s a trade off of a fan spinning—a fan cools but it also consumes energy. You have to run the fans at the right speeds so that they don’t pull more power than they produce useful cooling. Then we have workloads running on various processors, for instance on dual processors, if they are CPU intensive, then they tend to burn hotter. It’s the fine tuning of those algorithms that lets these servers reach the maximum efficiency.
Is it Intel only or are you also going to have AMD-based servers in Gen8?
These servers are based on the Xeon E5 but we will have the Interlagos family of CPUs that will also be introduced in the Project Voyager Gen8 families. What we are announcing today are the Intel-based solutions.
How does the data captured by the management chip work its way into dashboards?
The chip itself does the heavy lifting to monitor all the parameters and it packages the answers in industry standard protocols such as SNMP, SMASH, IPMI-based etc. that are consumable by other popular managers.
Have you built hooks into the facilities management software on one end and the hypervisors on the other?
Typically a VM doesn’t know much about the underlying hardware. What we do is that we know about the VMs and our management software offerings have the capability to move VMs to different physical machines based on utilization and power consumption. When it comes to interfacing with the facilities, we are also doing that. This generation is one of our more significant steps to doing that. We can now do physical placements of machines because location sensors happen to be built into them. The information as to where they are placed in a rack and row is available to the management software. If you have a facility situation where some aisles are hotter than others or airflow is sub-optimal then you can understand if you have placed your physical servers in a location in the data center that’s not correct for the amount of heat that it generates.
In these Gen8 ProLiant servers, we worked with Intel on something called Machine Check Architecture. If a VM goes awry because a memory cell goes bad in the VM memory space, we can quarantine that memory cell, kill that virtual machine off and continue to execute the existing farm of VMs on that subsystem. It’s part of the memory sub-system and the system software integration with popular hypervisors from VMware and Microsoft Hyper-V.
