HP is the world's leading server vendor, delivering solutions on both x86 and Itanium architectures and on Unix as well as Linux and Windows operating systems.
HP's Business Critical System (BCS) unit is the primary hub for HP's Itanium and Unix business, and it's a business unit that is growing to meet the needs of scale-up computing clients.
"In the high-end server group, we have a set of customers that are very demanding in their requirements in terms of mission-critical availability and reliability," HP Senior Vice President and General Manager of Business Critical Systems (BCS), Martin Fink, told ServerWatch. "While not always the case, a good chunk of our business is tied towards scale-up computing."
In scale-up computing, there is one single system image with lots of processors and memory. In contrast, with the scale-out model there are lots of interconnected servers. Fink explained that in the scale-out model, applications are often replicated many times over multiple servers and the work is distributed across multiple application servers.
Even though scale-up computing is based on a single server image, there are still multiple levels of load balancing and reliability present to distribute computing needs. Fink noted that from an infrastructure perspective, there are often many dual-redundant components.
"For example, with a dual cross bar fabric, all the CPUs as they scale and talk to each other have two fabrics," Fink said. "So if one goes down, the other can take over."
Another resiliency feature is something known as double chip kill. Fink explained that this feature allows a system to survive from double chip memory failures. There are also partioning options that can enable high resiliency. A system can be carved up into electrically isolated partitions within a single server to do things like failover as well as splitting up a workload across partitions.
Going a step further with high-availability and disaster recovery, servers can be clustered together. One server can be set up in standby mode, such that if any one of the servers in the cluster has an issue it would fail over to a backup server in the cluster. Redundant machines can also be connected across geographic distances as well, providing an additional layer of disaster recovery potential.
High Availability on Linux vs. Unix
Fink noted that that there are some similarities on the Linux side to the high-availability features available in Linux, though he added that in Linux they are not at the same level. Red Hat has clustering capabilities and there are also third party tools from Symantec as well.
"What they don't provide is the infrastructure resiliency, like double chip kill, for example," Fink said.
HP's effort for the next generation of BCS systems is wrapped up in the Project Odyssey banner. Fink noted that Odyssey includes ongoing continued investments in the HP-UX, Non-Stop and OpenVMS platforms.
There's also a second component to Odyssey that involves bringing similar capabilities to x86-based systems. One of the x86 components, codenamed DragonHawk, aims to bring Superdome Itanium, HP-UX type capabilities to x86 on Linux.
"What happens is people will sometimes have a tendency to focus on that (Dragonhawk) piece, as it's the new piece," Fink said.
One such focus on DragonHawk came in an interview with ServerWatch earlier this month. At the time Scott Farrand, vice president of Industry Standard Servers and Software in the Enterprise Group at HP, said that HP's go-forward strategy for mission-critical systems is shifting to an x86-based world.
Fink however stressed that Odyssey is about more than just x86.
"The reality is we very much want to continue to deliver our mission-critical capabilities in the other world," Fink said.
For the HP NonStop platform, Fink noted that HP still has two strong market segments, financial services and telcos. When it comes to HP-UX, Fink said that HP has an extremely large and loyal installed base of customers.
"We compete in every Unix opportunity in the market," Fink said. "There are still a number of customers that like scale up and Unix."
Selling Unix vs. Linux
When it comes to positioning Unix versus Linux, Fink stressed that he stays away from conversations on one operating system being better than the other. He noted that he tries to have conversations around scale up versus scale out, and in his experience different customers have different views on which approach works best.
As well, the decision is often made based on what skill sets are present at a particular customer deployment.
"It's really hard for me to give an answer, that in a particular situation I recommend Linux and scale out and in another I recommend Unix and scale up," Fink said. "In reality it's a conversation with the customer and it depends on the culture of the customer, environment, staff training and all sorts of things like that, so there is no one real right answer."