Upgrading to Sun SPARC Enterprise SMP Servers.

As vice president of Enterprise Systems in the Systems Group at Sun, Gary Beck is responsible for US-III and US-IV volume servers and datacenter servers, as well as the development of future datacenter systems. This position is the natural progression of a career that includes introducing some of the most successful UNIX products in history, such as the Sun SPARCcenter 2000, Sun SPARCserver 1000 and E3x00 and E6x00 systems.

Prior to being asked to lead the Enterprise Systems Group in 2004, Beck was vice president of the Integrated Products Group at Sun where he was responsible for the design, engineering, and marketing of a range of integrated products such as the Sun Reference Architectures, Customer Ready Systems, and Sun BluePprints.

Sun Inner Circle recently caught up with Beck, who has been hard at work on the recent launch of the new Sun SPARC Enterprise family of servers. This new hardware lineup includes upgrades to the Sun Fire line of SMP-based servers, as well as CMT-based CoolThreads servers.

According to Beck, Sun customers who rely on SMP-based hardware can now take advantage of numerous mainframe RAS (reliability, availability, serviceability) features for greater consolidation capabilities — and build a bridge between SPARC IV+ systems and the next generation of Rock/Supernova CMT servers Sun plans to release in 2009.

INNER CIRCLE (IC): What's the thinking behind upgrading from Sun Fire SMP servers to the new Sun SPARC Enterprise family?

BECK: Let me start by putting these servers into some perspective. Sun has a pretty long history of successfully providing its customers with consolidation features that significantly reduce cost, footprint, and environmentally adverse consequences of datacenter power consumption. With that said, the architecture of the new server family is an evolution of the existing Sun Fire line, which makes the transition fairly seamless. We think this is crucial to a lot of Sun customers who have a great deal invested in older SPARC/Solaris systems.

We want to make it easy for users to evolve their datacenters over time as we develop new systems. Just as it is an easy transition from the UltraSPARC IV+ Sun Fire servers, which are nearing the end of their technical lifespan to the SPARC Enterprise line, we expect the transition to the next generation of CMT-based servers in 2009 to be just as easy.

But in the short-term, users of older SPARC/Solaris systems are going to need new hardware. The UltraSPARC IV+ road map has seen an impressive range of scalability over the life of the servers. This range ends with the introduction of the 1.95/2.1 GHz processors earlier this year. So to provide large, SMP-based systems to a user base that depends on these kinds of servers, we've introduced the midrange M4000 and M5000 servers, along with the high-end M8000 and M9000 offerings. This powerful family is completed with making the CMT CoolThreads products part of the SPARC Enterprise line, offering our customers unmatched scalability, RAS, and price performance.

IC: What are some of the features that customers find attractive in the SMP members of the Sun SPARC Enterprise line of servers?

BECK: Right now we're getting a lot of positive feedback about increased performance and consolidation options. As far as performance goes, our customers are discovering that Sun SPARC Enterprise servers are faster than 1.8 GHz Sun Fire sServers by about 10 percent to 60 percent per core, and in some cases up to 100 percent% faster. In addition to faster processors, these servers have better aggregate bandwidth, I/O bandwidth, and larger memory — up to 2 TB of memory in the M9000, which is the high-end member of the new server family.

The new Sun Enterprise Server products include some significant new performance wins. First of all, we are excited to jointly announce with Fujitsu a new world record for the important SAP Standard Application SD benchmark. This benchmark represents critical tasks performed in real-world ERP business environments. Here the Sun SPARC Enterprise M8000 set the record for servers with 24 or fewer processors by beating IBM System p5 570 by 32% and HP Integrity Superdome by 30 percent.

The Sun SPARC Enterprise M9000 achieved over a teraflop on the Linpack, out-performing the best-published single system from IBM System p5 595 (1.9 GHz POWER5) by over 2X and passing HP's Itanium 2 processor-based Superdome (1.6 GHz/24 MB) by nearly 29 percent. Finally, the SPARC Enterprise M9000 set new world records on the SPECompL2001 and SPECompLbase2001 benchmarks, again passing IBM and HP competitive systems by 16 percent and 22 percent%, respectively.

These performance features are exciting, but what's arguably even more interesting can be found in how we're approaching consolidation with a whole slew of mainframe RAS features — such as high levels of availability and manageability — without the high cost, complexity, and vendor lock-in typical of big iron. We've also added mainframe-class reliability features that are really important for business continuity. Things like instruction-level retry, protected SRAMs and registers, memory chipkill, and mirroring — as well as end-to-end ECC protection, hot-swappable components, and hardware redundancy — are all part of the SPARC Enterprise family of servers and are offered at open systems prices.

When you add Sun virtualization technologies like dynamic system domains and Solaris 10 containers, you get an extraordinarily resilient and flexible UNIX consolidation platform that Sun/SPARC users can leverage immediately.

IC: Do these new RAS features reduce the level of complexity that is often associated with consolidation efforts?

BECK: The ability to consolidate largely depends on server components and how well these components can be serviced and replaced without bringing the entire system down. This is an area, of course, where mainframes have traditionally done very well, and this is the reason why we've incorporated dynamic resource management into the SMP members of the SPARC Enterprise server family. Dynamic resource management lets you take a failing component out of service without bringing everything to a grinding halt. This is pretty hard to achieve, by the way, in UNIX servers from other vendors. These products typically have to be turned off if there is a hardware issue, whether it's taking care of a problem or consolidating servers.

IC: So is virtualization approached differently in the Sun SPARC Enterprise line than in UNIX servers from other vendors?

BECK: All hardware vendors can virtualize their systems, of course. But how they go about partitioning large numbers of processors and memory into smaller granular pieces for running workloads varies. Sun's dynamic system domains — in both hardware and software partitions — along with Solaris containers allow a combination of up to 24 hardware partitions and thousands of Solaris containers. The result is very low overhead.

Virtualizing with competing UNIX systems requires much more overhead. And if you go beyond a certain number of partitions with these servers, the rate of return can diminish significantly. In fact, you can actually bring these other systems to their knees because other virtualization approaches typically reach a point where server utilization is consuming more overhead than doing useful work.

IC: It sounds like all four members of the SMP-based SPARC Enterprise server family share similarities in providing highly scalable virtualization and consolidation environments. But how do these four classes of servers differ?

BECK: Each server is extraordinarily scalable. You can go from a single-socket system based on dual-core SPARC64 IV processors to up to 64 sockets with 256 threads. And each of the members of the line are meant to provide the sort of rock-solid reliability that customers of older Sun Fire V and E class servers need for mission-critical business applications.

The differences essentially boil down to memory and the number of system domains. At the very high end, the M9000 provides up to 2 TB of system memory, which is the highest Sun has ever offered. The other high-end server in the family, the M8000, provides up to 512 GB. The midrange systems made up of the M5000 and the M4000 respectively provide up to 256 GB and 128 GB of system memory ECC. Similarly, the number of system domains ranges from the M4000 system's two-system domains all the way up to the 24 domains found in the very high-end M9000 servers.

IC: How does the operating system help with consolidation — and long-term investment protection?

BECK: Solaris 10 is a huge differentiator with these servers. The operating system ensures that the transition to SPARC Enterprise hardware means no recompiling, which makes the change pretty seamless. It's all part of the Solaris binary compatibility guarantee; you install it, integrate it, migrate the workload — and then run. It doesn't matter which version of Solaris you've been using. This provides significant investment protection by allowing our customers to choose other platforms and vendors if they wish, because Solaris runs on multiple hardware architectures.

Other advantages to having Solaris installed in these systems include more features than provided by other hardware vendors, including things like DTrace and performance optimization features, ZFS, and Sun Fault Management Architecture (FMA). You also get security built right into Solaris in a way that just doesn't happen with Linux and AIX.

IC: How do the new RAS features help with business continuity?

BECK: RAS business continuity features are a big part of the Sun SPARC Enterprise line. This line of servers has a number of RAS enhancements you would expect to find in mainframes. There are things like instruction retry, data check pointing, memory chip kill, memory mirroring, and memory redundancy. There is also automatic diagnosis and recovery to simplify operations when errors do happen. And again, this new line of SMP servers comes with Sun FMA.

This feature analyzes the thousands of different error messages Solaris produces about any kind of hardware failure in the server, whether the error message is chip-related or pertains to something like a disk drive problem. FMA shows which problems are critical — and which problems can be less critical — and suggests what to do about these situations.

IC: You've mentioned some benefits of using Sun SPARC Enterprise servers as compared to offerings from other vendors. How specifically do the new Sun servers stack up against these competing products?

BECK: Solaris Containers are a big part of the equation, because they provide our customers with more consolidation opportunities. Let me give you an example. The M4000, which is a midrange member of the Sun SPARC Enterprise family, supports two dynamic system domains, or partitions, and each system supports up to 8,000 Solaris 10 containers. Now compare the M4000 with the high-end IBM p595, which has no hardware partitions whatsoever and only supports 250 software containers.

And once again, the operating system is a big differentiator. Both the IBM System p5 595 and HP Superdome 32 need one instance of an operating system for each software partition while any member of the Sun SPARC Enterprise family needs only one instance of Solaris for thousands of containers. And when you consider price, there's really no comparison, because Solaris containers are free, but HP's vPARS and IBM's LPARS and mPARS carry significant cost.

IC: How do the consolidation and virtualization features of the SPARC Enterprise server line help decrease energy bills and reduce the need for more datacenter floor space?

BECK: By consolidating dozens or even hundreds of older servers onto one or a few of the new SPARC Enterprise servers using the virtualization features we've talked about — dynamic system domains and Solaris containers — you can save considerably on power and cooling bills, as well as free up space in the datacenter. These savings have been well documented for the T1000 and T2000 CoolThreads servers, the predecessors to the M1000 and M2000. But the same concepts apply to the M4000-M9000 systems as well.

IC: We started this conversation by asking about the rationale for upgrading to the Sun SPARC Enterprise line of SMP servers. Perhaps a good way to conclude would be to ask how Sun is helping customers with the transition.

BECK: I should probably point out that the transition can be from older Sun Fire models just as easily as it can be from HP and IBM systems. We offer the Sun Technology Refresh Plan and the Sun Upgrade Advantage Program to help customers move from older Sun platforms or systems from HP or IBM. We've also introduced a number of very creative purchasing plans, including leasing options.

These offerings provide financial relief for following the Sun product road map, while also making sure that enterprises can get significant performance, power, consolidation, and environmental benefits without breaking the bank.