x86 servers have long been associated with the low end of the server performance totem pole. But developments, such as multicore chips and enhanced memory, have made the average x86 box into a formidable machine. HP even aims its ProLiant Scalable System or SL line at the high-performance computing (HPC) market. “We have seen a market shift as HPC has become more affordable and available to users,” said Glenn Keels, director of marketing, Hyperscale Business Unit at HP. “The typical obstacle is based on affordability, power efficiency and complexity, and we are once again crossing that barrier to even greater levels of performance that will foster increased innovation.”
No longer just for the low end of the server performance totem pole, x86 servers
are increasingly finding their way into the HPC market. With its ProLiant Scalable System line, HP has made great strides to fill this
Perhaps 10 years ago, HPC was a very clearly defined niche. These days, the typical desktop or laptop could almost be considered HPC by older definitions. So how is HPC being defined these days, and how do these machines differentiate themselves from standard servers and other ProLiant models?
“The performance bar is always raising,” said Keels. “The supercomputers of yesteryear are becoming mainstream today.”
Every decade or so, he said, HPC makes a step change, thanks largely to improvements in performance of processors, systems, interconnects, and so on. Keels’ view is that the current barrier of affordability, power efficiency and complexity is being addressed –- in this case by HP (NYSE: HPQ), through the use of accelerators, graphics processing units (GPUs), cluster solutions, and other bells and whistles, to soup up its regular ProLiant boxes to achieve a hike in performance per dollar, per watt and per square foot.
HPC History Lesson
Back in 1997, the first-ever supercomputer at Sandia National Labs broke the 1-teraflop performance barrier. Similar performance can now be gotten from a single HP ProLiant SL390s server with a far smaller footprint. To view the leap in the HPC landscape another way: Keels pointed out that in 2002 the No. 1 system was Earth Simulator in Japan, yet it had been surpassed by a single rack of the SL390s in the Tsubame2 cluster at Tokyo Tech. This rack is 2000x smaller, uses 200x less power, and it is available at 450x lower cost.
Keels knocks the approach taken by competitors. He said they either create new, costly, complicated, proprietary technology suited to a narrow problem or throw massive amounts of servers, processors, cables, energy and real-estate at the problem. The result is high cost, complexity and ferocious quantities of energy consumption.
With the SL series, HP has taken a different tack. It integrates purpose-built HPC compute, storage, management software and networking, as well as power and cooling infrastructure, all based on a standard x86 architecture.
Keels provides an example of the SL390 as the core of such a converged HPC solution. Tokyo Institute of Technology’s HP-based Tsubame2 supercomputer is ranked in the top 1 percent of supercomputers worldwide.
“It’s more than 20 percent more affordable, has twice the performance per server, and 50 percent greater performance per watt than its next competitor,” said Keels. “It’s the greenest production supercomputer on the planet.”
Proliant Scalable System Rundown
So what does the SL line consist of? There are nine main models with a few variants. Four of them are brand new: SL160s, SL165s, SL335s and SL390s.
The entire SL range can be split into two main camps, each with several models.
- SL6000 System
- ProLiant SL160z G6 Server series
- HP ProLiant SL165z G7 Server series
- HP ProLiant SL170z G6 Server series
- HP ProLiant SL2x170z G6 Server series
- SL6500 System
- HP ProLiant SL160s G6 (NEW)
- HP ProLiant SL165s G7 (NEW)
- HP ProLiant SL170s G6
- HP ProLiant SL335s G7 (NEW)
- HP ProLiant SL390s G7 Server series
- HP ProLiant SL390s G7-1U half width server
- HP ProLiant SL390s G7- 2U half width server
- HP ProLiant SL390s G7- 4U half width server (NEW)
For the purposes of this article, we will focus on the four newer models.
The HP ProLiant SL160s G6 Server is the twin of the regular ProLiant DL160, except it has the added SL6500 infrastructure to turn it into what Keels described as the “workhorse of hyperscale data centers, spanning both edge and mid-tier web applications as well as HPC.”
It comes with up to 2 Intel Xeon 5600 series (using the Intel 5520 chipset), up to 192 GB of RAM, an embedded HP Smart Array, a B110i SATA RAID controller, and 6 normal sized SATA/SAS disks or 8 small form factor (SFF) SATA/SAS/Solid State Drives (SSD). Four of these servers fit in a 4U chassis.
G7 is the twin of the ProLiant DL165. Applicable for many web application
tiers, it features the AMD Opeteron 6100 series processors, which can cope
with demanding and memory-intensive HPC workloads –- up to 288 GB of RAM,
with the same storage specs and size as the SL160s.
The HP ProLiant SL335s G7 features up to two AMD Opteron 4100 series processors and 128 GB RAM. It comes in a half-width form factor. It is slated for dedicated hosting and web workloads that are not compute-intensive. On the storage side, you get four regular-sized SATA/SAS or 8 SFF SATA/SAS/SSDs. Eight servers fit into 4U of an SL6500 chassis.
The new HP ProLiant SL390s G7 4U is the latest model in the SL390 family. The SL390s is a 4U set up that features an integrated GPU design that can handle up to eight GPUs in one server. Two SL390s can fit in a 4U chassis, which equates to up to 16 GPUs in one SL6500 series chassis. This model is aimed at heavy-duty GPU applications and HPC sectors, such as oil and gas, life sciences, financial services and education. It comes with up to two Intel Xeon 5600 Series processors, up to 192 GB, an HP Smart Array, a B110i SATA RAID controller and eight SFF HP SATA/SAS drives.
“The main competitors to the SL6500 System are more traditional approaches to hyperscale computing, including general-purpose rack servers and custom servers in the web space, and proprietary architectures and GPU-connected general purpose rack servers in the HPC segment,” said Keels. “In both areas, the SL6500 System provides a common modular architecture that allows optimized performance, energy efficiency, density and simplicity for any application at any scale.”
Which models within this line are the most popular with users? In the web space, Keels said, the SL160, SL165 and SL170 are the most in demand. In the HPC space, the SL390 servers as well as the SL165 tend to be the most popular.
“As a new entrant, the SL335, with the balanced affordability, performance, energy efficiency and density is quickly gaining the attention of not only web but also dedicated hosting providers,” said Keels.
Drew Robb is a freelance writer specializing in technology and engineering. Currently living in California, he is originally from Scotland, where he received a degree in geology and geography from the University of Strathclyde. He is the author of Server Disk Management in a Windows Environment (CRC Press).