One of the challenges in a Mission Critical environment is managing application response time. In one of my past roles, I was a product manager for HP Insight Dynamics - VSE Workload Management for HP Integrity Servers (a.k.a. Global Workload Manager), HP-UX Workload Manager, and HP Secure Resource Partitions (a.k.a. Process Resource Manager), now all part of HP Insight Dynamics - VSE.
One of the things that we worked on, particularly with the workload managers, was assigning enough resources to each workload to maintain the appropriate service levels. This turned out to be a lot of work, especially when you consider how many applications run on HP-UX 11i v3. We found from an HP-UX 11i and server perspective that it was easier to instrument CPU utilization as a proxy for application performance. It turns out that response times were usually consistent until CPU utilization was above 95%. If we conservatively set the CPU utilization threshold at 80%, and then add additional CPU resources, the application normally had a consistent response time.
Well, IBM has
announced their first Power 7 servers. They, like they have done in the past,
focused on a few key messages: performance, power efficiency, and system
management. They added details in a few other areas as well, but nothing
drastic beyond a more powerful processor and therefore more powerful systems.
Since Intel has
announced the Itanium 9300 processor (Tukwila), but HP hasn't announced its
server line up with these processors yet, doing a head to head hardware
comparison isn't appropriate at this time.
that is shipping today is the UNIX operating system that is supported on each
platform. Power 7 servers support AIX 6.1, and HP-UX 11i v3 is supported on
current and future systems.
While I've heard
more than once that people thought HP-UX 11i was dead, indeed, it is still
alive and well. Thanks to the work done by Gabriel Consulting, we have a good
idea of what customers value in each UNIX operating system.
From a HP-UX
11i v3 perspective, particularly when comparing to AIX, a few things pop
HP-UX 11i customers
tend to think more highly of HP's virtualization technology, particularly when
you include multi-system virtualization capabilities and management (for
Instant Capacity and Insight-Dynamics
- VSE). I wonder if that is perhaps because HP Integrity Servers and HP-UX
11i v3 offer multiple types of virtualization technologies, so customers can
pick the technologies that work best in their environment, and yet use the same
management console for any and all virtualization technologies?
This leads directly
to the second thing customers really appreciated about HP-UX 11i - the
simplified, single-pane-of-glass management and management automation. Tools
Systems Insight Manager allow the management of physical and virtual
environments. IBM Director appears to offer similar capabilities, but the last
time I checked (and it's been a while), you often get to switch between
different tools (with potentially different log ins, etc.). HP has done the
hard work to truly integrate many of these products, and perhaps customers
actually appreciate it.
The idea of integration continues to the next
reason customers prefer HP-UX 11i - integrated high availability, disaster
tolerance, and virtualization. Not only do all of these products work together
on HP-UX 11i v3, but HP offers a lot of application integration and support.
This includes scripts from the Enterprise
Master Cluster Toolkit, Serviceguard Extensions for SAP or Oracle, and
Insight Dynamics - VSE Reference Architectures that show you how to build
everything together and get it to work - quickly, easily, and with fewer
Overall, it appears that HP customers value the integration that HP designs
into HP-UX 11i v3 - whether the operating environments that simplify ordering
and license management, integrating add on products such as virtualization and
high availability, or the information on how to deploy it with common
applications much easier.
So, if you use HP-UX
11i, why do you like it? Any of the reasons above? Or do you have other reasons
for prefering HP-UX 11i?
Over the years, I've met with many customers who have spikes in their holiday traffic. I've spoken with a southern hemisphere beverage company, who has a huge spike in orders the last Monday morning before Christmas. I've spoken with a customer who's busiest day of the year is the final Friday before Christmas. I've spoken to numerous retailers who have their busiest shopping days at this time of the year. Often, these spikes are 10 times or more higher than the average demand.
How do these customers adapt to the high levels of seasonal demand? The first, and most obvious technical way is to provision their systems to handle the peak demand. Of course, that means they are paying for excess capacity for the rest of the year. Having said that, they meet their business requirements, customers are happy, and the IT department keeps their jobs.
The alternative of reducing the peak size of the systems, so that they can't handle all the demand, will save a little money on the IT budget. However, every year, there are IT infrastructures that get a surge of demand that they weren't designed to handle, and the company ends up losing customers, their reputation, and a lot more money than the extra capacity would have cost in the first place.
Having said that, more and more customers are looking at this environment, and with reduced budgets, they want the best of both worlds. Customers need to handle their peak capacity, but also take advantage of lower costs. At the end of the day, there are two ways that virtualization can help in this situation.
First, and perhaps the easiest way, its to take advantage of some sort of flexible financing so that you only pay for additional capacity when you actually need it. This is the idea behind offerings such as Instant Capacity and Temporary Instant Capacity on HP Integrity servers all the way to truly flexible cloud computing offerings such as Amazon EC2.
The second way is to run additional workloads on the systems to use up the extra capacity. This works well, as long as those additional workloads can be released to provide resources for the primary workload when the demand spikes come along. Dynamic hard partitions (nPars), dynamic vPars, virtual machine, and application stacking technologies all make this possible . Freeing up resources can be everything from manually shutting down low priority workloads to automatically shifting resources between partitions to migrating workloads off of a system. I've even come across some unique ways of tackling this problem:
- locking down the environment for a few months, and shutting off all development and test systems;
- running on a single node of Oracle RAC for most of the year and expanding to multiple nodes for the holiday rush;
- migrating production workloads to larger or dedicated systems for a period of time
- and more.
The good news is that virtualization technologies, such as Insight Dynamics - VSE , whether on HP Integrity servers, HP ProLiant servers, or HP BladeSystem, create an environment where this is not only possible, but relatively easy to do.
Actually, these customers have it relatively easy. They know that they will have a holiday spike. They even can generate a reasonably accurate estimate of the workload that their systems will see on those days. They can plan to lock down their environment in advance to free up test or development systems. They can manually resize partitions days or weeks in advance. And since the holiday season is reasonably predictable, they can make there plans well in advance.
The nice thing about Insight Dynamics - VSE for HP Integrity is that while it makes it easier to handle the predicted fluctuations, it actually excels in handling the unpredictable spikes and troughs in demand equally well. Since it is automated, tools like the HP Global Workload Manager component in Insight Dynamics - VSE for Integrity can observe and react to changes in the environment in seconds- not minutes or hours. It automates the rest of the portfolio, including the partitioning, clustering, and instant capacity products to automatically react to changing workloads.
At the end of the day, automation of a flexible environment provides the best of both worlds - high levels of utilization (and therefore lower total cost of ownership), but with the ability to handle peak workloads - whether predictable peaks like the holidays, or an unpredictable peak. The best of both worlds - and a less stressful holiday season for all those who work in IT.