Reality Check: Server Insights : scale-out

I received a new HPC Multi-core server today – Intro to power measurement

d-field — Thu, 06 Nov 2008 00:02:00 GMT

Until a couple of years ago, when we referred to performance measurement of an application, we meant the amount of time that it took the job to run vs. the specific resources it used - number of cores, number of servers if you are using a cluster, the specific characteristics of the server cores and memory and other server specs, plus IO/storage resources and specs.

Basically, we only measured one thing, the elapsed time of the job. Then, using the resources and specs, we computed lots of things - throughput efficiency, parallel scalability and efficiency, performance per core or per server, IO metrics, etc, etc.

Now, we make an additional measurement - power utilization, which we correlate in time with the execution of an application. We want to know the average power used during the execution of a single job, and we also look at the variation of power during a job, and the maximum power used.

Of course, lots of people measure power used by computers. But, since most of these people are system managers or system designers, they don't have a reason to correlate power with specific applications and compute jobs. They want to know the average and peak power used to run their overall workload, so they can plan for current and future power requirements. This is important work, but it does not give them the ability to optimize their workload.

If you measure the average power used during the execution of one compute job, and you multiply that power by the elapsed time of the job, you have Application Energy - the electrical energy used to run that specific job. This is a very convenient quantity, since it gives you a single number that relates power usage to compute jobs. You can use Application Energy to optimize your workload, just as you use elapsed time.

A couple of examples:

1. You can measure application energy for a given set of applications on two or more different server models, and then select the more energy-efficient model. You can use this App Energy comparison together with elapsed time comparison, and then make speed vs. energy tradeoffs. If a job runs 30% faster but consumes 50% more Application Energy on Server A than it does on Server B, which is a better choice for your requirements?

2. We are also using Application Energy to determine the most efficient way to run applications which run in parallel on a cluster of servers - a common way to run HPC codes. For one common HPC application, we ran the same job at 3 levels of parallelization and compared the elapsed times and Application Energy. We showed that the job used only 4% more Application Energy running 32-way-parallel (on 32 cores) vs. 16-way parallel. But the job used 20% more Application Energy running 64-way-parallel vs. 32-way-parallel. In other words, there is very little energy cost using 32 cores and returning the results to the user much faster vs. using 16 cores. But there is a substantial energy cost to use 64 cores, which returns the results even faster.

Does anyone find this interesting, or agree (or disagree) with this approach?

I received a new HPC Multi-core server– Learning from standard benchmarks

d-field — Wed, 15 Oct 2008 22:07:00 GMT

Now that we have configured the hardware components and firmware settings in a known and hopefully optimum way, it is time to run the 1^st performance test. Personally, I like to run the memory bandwidth benchmark lmbench first, since there is a lot to learn from it. This benchmark computes the time required to move different amounts of data from the caches and memory to a core.

A modern server has multiple levels of caches - 2 or 3 levels. The highest level may be shared among several cores. The output of lmbench shows plateaus for each level of cache, showing both the latency to the cache and also the size of the cache. If the system has NUMA memory organization, then lmbench shows the latency for each of the NUMA "hops", as the data traverses the system topology.

Usually, the latency of each cache level is a fixed number of processor clock cycles. It is both interesting and important to know this number. It's interesting, because it allows you to compare different cache architectures, even if the systems you are comparing have different clock speeds. For example, it is interesting to me that the 1^st level cache latency of several modern servers, with very different architectures, is 3 cycles.

And it is important, because you are not always sure what clock speed your system is running, and you can compute it using lmbench. Or, you might encounter the problem we hit yesterday - on a 2-processor pre-production server, the two processors were running at 2 different clock speeds! This is of course not good - a user observed strange performance, but lmbench took the mystery out of the problem and told us exactly what was happening.

You can run lmbench in different ways, and each provides additional understanding.

The "stride" method runs sequentially through memory addresses and provides best-case latency.

The "random" method accesses memory addresses randomly, showing the worst-case latency. It is very useful to know the best-case and worst-case latencies - if their ratio is small, then memory performance is somewhat predictable. If the ratio is large, prediction becomes difficult.

If you set lmbench to access memory in units of the VM page size, then you can observe the impact of TLB misses on latency.

And, if you run lmbench simultaneously on cores which share a cache, you learn about the behavior of the shared cache.

Memory latency is a great tool if you need to unravel the architecture of a new server!

HP Tops TBR Customer Satisfaction Survey

Gordon Casey — Wed, 20 Aug 2008 19:33:00 GMT

Technology Business Research (TBR) recently released its 2Q-2008 x86 Server Customer Satisfaction Study results, and the news is EXCELLENT for HP this quarter.

HP ProLiant achieved the #1 ranking in both the Weighted Satisfaction Index (WSI) and the Overall Satisfaction Rating. The WSI ranking is a weighted customer satisfaction score that considers what is important and how satisfied customers are with each area of customer satisfaction. The overall satisfaction rating is the response from customers when asked “Overall how satisfied are you?”. Together, these two indicators provide us with a quick snapshot of customer satisfaction for each vendor surveyed.

With these results, HP now has sole leadership and no longer shares the #1 position with the competition. HP was the only competitor to demonstrate gains in every category measured by TBR (9 in total). Specifically, the results in Hardware Quality/Reliability, Management Tools, Server Value and Parts Availability were significantly higher than industry averages. TBR identified these four (4) areas as providing a competitive advantage for HP over its competition. HP also scored higher than industry averages in the all-important category of Overall Satisfaction. According to TBR, customer expectations continue to rise as customers seek out new technology advances to counter serious datacenter challenges. Customers are also searching for much more than attractive purchase pricing. Server value incorporates all expectations, incorporating everything from reliability to technical support and management tools. The influence of relationship quality also figures strongly into customers’ perceptions of value. Our focus on delivering the products customers want continues to pay off. For nine (9) consecutive reporting periods, HP has remained ahead of the competition as the brand perceived as most differentiated. Customers continue to recognize the value HP delivers in the ProLiant brand.

And it gets better! Customer Loyalty is another excellent category for HP. For the last seven (7) quarters, HP has held the #1 position in customer loyalty. HP scores in customer loyalty for the last two quarters are the highest scores on record (since 2Q-2005).

One does not have to look far when looking for reasons why HP dominates customer loyalty. Customers recognize differentiated value. One example is in the area of server management. TBR reports that HP customers place management higher in priority than customers of our competitors. Customers expect more from HP and the results show HP has delivered. HP Systems Insight Manager (HP SIM) is a single software tool providing a consolidated view of everything you need to manage your server and it comes with every HP server. TBR also measures customer expectations, customer priorities, and how well a company is meeting those needs. According to the survey, customers consider hardware quality, parts availability, and overall value as the three most important attributes of a server vendor. For customer expectations, scores show HP is meeting customer expectations in every measurable category, and, more importantly, HP ranks #1 in all three areas. HP is focused on these attributes and also sees them as critical to our customers’ success. We continue to drive improvements in these areas by working with our developers, partners, and suppliers to drive quality into every server. As an example, by designing servers focused on customer self repair (CSR) customers have the flexibility to schedule and make a repair at their convenience. Where do we go next? These results have set the bar higher and clearly identified the market leader. Our challenge moving forward will be to accelerate the pace and deliver on the expectations of our customers. We’re committed to doing this – and it is evidenced in our offerings from the last few quarters. HP is listening to customers and delivering servers that reduce power consumption, expands virtualization, and enable success in the most demanding and complex computing environments. I welcome your comments – tell me how we are doing. Are there areas we can improve? Let me know – HP’s focus is on the customer and we need your input.

Views of the HP POD

aimeeschoaf — Fri, 18 Jul 2008 18:35:00 GMT

The HP POD had a very busy week indeed... and we had some great photos taken while it was in the Silicon Valley. Here are a few of those pics - thank you for participating! For more information on the HP POD, go to: www.hp.com/go/pod.

Look for the HP POD this Fall at an industry event near you!

The HP POD finds Yahoo... Photo by Ryan Knight

The HP POD introduces itself to Google -- photo by Holly Benson

Photo by Cheryl Edwards

Will Hayes says, "Wait, come back! You forgot to drop off my TV!

Photo by Carlos Mangandy

Congratulations to the winner of our "Spot the POD" Drawing!

aimeeschoaf — Fri, 18 Jul 2008 17:18:00 GMT

Congratulations to Kenn Durrence, who was drawn as the winner of our "Spot the POD" Drawing. Kenn captured the HP POD in San Jose.

Congratulations, Kenn! Please email aimee.schoaf@hp.com with your contact information!

Thank you all for participating in this promotion - watch for the HP POD coming your way again this Fall.

“Spot the POD” Drawing

aimeeschoaf — Wed, 16 Jul 2008 06:22:00 GMT

The "Spot the POD" Drawing is now closed. We will be randomly choosing one winner from the eligible entries on Friday, July 18, 2008. Here are some of the photos posted as the POD made its journey through Silicon Valley! Thank you all for participating! For more information on the HP POD, go to www.hp.com/go/pod.

Nascent

aimeeschoaf — Wed, 16 Jul 2008 06:13:00 GMT

By Christian Post

"Nascent: coming or recently having come into existence."

It would be a short argument to suggest that containerized datacenters are nascent. They have been developed and deployed for years to serve mobile, military, and disaster response needs. Yet we are seeing a ground-swell of containerized offerings at the convergence of unprecedented energy costs, a resurgence of compute demand, and the growth of business dependence on IT. One could claim the interest in containerized datacenter solutions has now emerged beyond background noise and appears to be forming a "nascent" marketplace.

Who is paying attention? Certainly enterprise scaleout, Web 2.0 and HPC customers. Containers offer shorter time to provision and noticeable capex and opex savings. More than clamoring for provider briefings, some users are developing their own concepts. Further, mid-level managers see savings beyond the 20% barrier that normally hinders proposing new approaches to executives. The datacenter construction industry is certainly taking notice: Here is a diversification opportunity in which to invest, a competitive threat to take notice, and a potential fad to counter. Container integrators, also, note the opportunity and are seeking partnerships to extend their reach beyond non-IT applications. Finally, IT providers are staking claims. To them, containers extend the sales and solution opportunity beyond the box, enclosure, and rack to a sales opportunity valued at 2000+ nodes per order and solution opportunity in which to differentiate. How long will it last? That's the problem with "nascent." The early adopter era nears and likely precedes the pause. In the pause, investors and watchers will reflect on whether the return merits the convergence of facilities and IT and the new processes in deployment and operations.

So who will be the preferred provider(s)? Consider the following criteria: First, containers bring together technology at the MEP&IT spheres. Certainly design leadership that offers density, cost, and deployment prowess are mandatory. Second, IT configuration flexibility will be imperative, especially to those seeking best-in-class performance per watt or heterogeneous flexibility. Third, unique services for deployment, break/fix, and management of containers must be included. Few customers would leave any of these three out. But one is still missing. As varied as there are datacenter strategies and building designs, few customers express exactly the same container design expectations.

Thus there is a fourth criteria. In a new market, the provider that can 1) offer sound advice on whether to and how to deploy containerized datacenters and 2) respond to container design modifications quickly with skill and reputation, will likely emerge most trusted for such a "nascent" investment. What do you think? What is "nascent" about the containerized datacenter market? How long will it last? What kind of provider best serves the customer need?