Advanced Computing in the Age of AI | Thursday, March 28, 2024

New Dell GM Talks System Trends 

It has been a year now since Dell took itself private and removed itself from the prying eyes of Wall Street investors. The company had its share of drama as it went private, and its main rivals in the systems business, Hewlett-Packard and IBM, are now going through their own divestitures and related theatrics. Now, Dell talks about its business when it wants to and how it wants to. Ahead of the annual Dell World conference held in Austin, Ashley Gorakhpurwalla, general manager of server solutions at Dell, sat down with EnterpriseTech in his first interview after taking over that role a few months back to give us some insight into Dell's system business.

Every little bit helps. Dell issued a Fiscal Year 2015 State of the Business report last week, and now that Dell is privately held, the company did not provide much in the way of shipment and revenue figures for its systems business. For its Enterprise Solutions group, which sells servers, storage, switches, and systems software, the company did cite IDC market data that showed Dell having the number one market share in the second calendar quarter in servers in North America at 24.8 percent unit shipment share, and was number one in Asia/Pacific-Japan with 19 percent share; Dell was behind rival HP in EMEA with 22.1 percent share of shipments in the quarter. The only other interesting piece of data is that Dell's channel has brought in more than 4,300 net-new customers for its enterprise products with over 10,000 new orders between February and August. This is not very much in the way of information.

In his new post, Gorakhpurwalla is not permitted to provide much in the way of detailed financial information about the Dell systems business, but he can talk a bit about trends that are driving the business.

For those of you not familiar with Gorakhpurwalla, he joined Dell in 1999 from Compaq, just when Dell was working on its third generation of PowerEdge X86 servers. Fifteen years and ten server generations later – yup, that's Moore's Law right there – Gorakhpurwalla has worked solely in the Dell server division with a focus on engineering. This is a bit unusual at Dell and other IT giants in that top executives tend to train across different lines of businesses and with different kinds of responsibilities.

When Forrest Norrod was tapped to start the Data Center Solutions (DCS) custom hyperscale server business a few years back, Gorakhpurwalla was put in charge of the general-purpose PowerEdge line. DCS quickly grew to more than $1 billion in sales. Two years ago, Dell merged DCS back into the PowerEdge business, with Norrod in charge of the whole shebang, and when Norrod left Dell a few weeks ago to join chip maker AMD to head up its server and embedded systems chip efforts, Gorakhpurwalla was tapped to run the entire systems business at Dell.

One of the reasons why Dell merged DCS back into the core PowerEdge business was that the lines were blurring between custom machines and general purpose machines, and intentionally so with the advent of the so-called PowerEdge-C systems, which were a commercialization of some of the ideas that came from custom server engagements with hyperscale datacenter operators that were intended to be sold to a broader base of customers that wanted minimalist, dense machinery with no frills.

"There are some customers that we do not classify as the standard hyperscale folks but they do buy hyperscale products from us," explains Gorakhpurwalla. "And just the opposite, there are some folks that have tremendous scale but really just want to buy the mainstream PowerEdge products because they want a certain level of management, a certain level of capability, global reach, or something like that. We continue to be very successful across all of those spaces. The DCS space changes – it is pretty dynamic based on customer needs – and it is still a $1 billion-plus business if we would carve it out separately, which we do not."

The dynamism of DCS is not just in the server engineering, but in the choppiness of huge orders, which come and go in waves and are not a steady revenue stream at all. Moreover, says Gorakhpurwalla, hyperscale customers in the United States want different things compared to hyperscale customers in China.

"The classic PowerEdge space continues to grow for us, and year-over-year in the double digits," added Gorakhpurwalla. That is about all the detail we are going to get on revenues.

The last financial figures we have seen out of Dell were for the second quarter of fiscal 2014 ended on August 2, 2013, and in that quarter Dell posted $2.89 billion in servers, peripherals, and networking revenues and $432 million in storage. At the time, the server portion (which again included networking and peripherals other than storage) was growing at 24 percent. In the rolling four quarters from July 2012 through August 2013, Dell had $10.5 billion in server and related revenues and just under $1.7 billion in storage sales, which was shrinking slightly as a partnership with EMC was still winding down.

The impending launch of the "Haswell" Xeon E5 processors from Intel did not have much of a downward effect on Dell's server sales in the spring and summer of this year, which was a bit unusual.

"There is usually a trough before a major technology launch that," says Gorakhpurwalla. "The last ten launches I have been through at Dell, we typically see that some of the market will stall, while other parts that pick up because customers have standardized on a specific product and business continuity is more important than the performance bump they might receive. It usually balances out, and then depending on the benefits perceived from the newer technology, that shapes the curve a little bit." The "Nehalem" launch in early 2009, for instance, had a huge pause in it caused not just by the substantial performance improvement that was coming with these Xeon 5500 processors, but also because the Great Recession was in full roar in 2008. "This time, we haven’t seen that stall."

Part of the reason why this is the case is that a processor launch is becoming less and less about the CPU performance gains and more about having a natural point where customers can adopt a whole different slew of technologies for their systems.

"We like to make sure we are differentiated and we offer the highest performance, but for us, with the Haswell generation, it was more about customers wanting to move storage and networking closer to the processor and adding in technologies that help more customers move towards scale-out," Gorakhpurwalla says. "There are some hyperscale customers that have tuned their software to a specific number of cores because they want a specific ratio of cores to storage to networking, but those folks have been heading down that path for a while and working with us and Intel to tune for that. There are some customers that really do buy the highest core count, and their software has the capability to utilize it. Although customers do like to standardize, they still do buy different things for different purposes. And we see others that want to buy in the sweet spot of performance or price because they are tuning their cost of goods sold, and others are moving up the SKUs because they can write their software to take advantage of the cores. If you look at the trend over the past seven years, there has absolutely been an acceptance of higher core counts."

With the Haswell launch, Intel has 39 official parts and over 20 customized parts that it has created for various customers (presumably most of them are hyperscale or HPC shops with very specific needs). At the very top end, the company has a 16-core and an 18-core variant, which offer the highest core count in the server market at this point. Intel has not provided prices on these units, so it is hard to reckon what kind of price/performance they deliver, and thus far, many system makers have been topping out with the 14-core Haswell parts in their systems.

Most hyperscale companies control their own software stacks, but in the commercial HPC arena, companies tend to license their software and the decision on what processor to buy often has more to do with the nature of the software license than the feeds and speeds of the processors available in a PowerEdge system at the time.

"In HPC, it depends on what type of software they are running," Gorakhpurwalla explains. "If it is something where they are paying for software licenses, those licenses may be the pivot. They want as many cores and performance that will fit under a license. We have been in interesting discussions with customers who would be willing to pay for a server that can run overclocked for a year – as long it didn't give incorrect results – because it helps them avoid buying more software licenses."

This, of course, sounds exactly like the kind of overclocking that goes on in some financial trading systems. But every industry has their way of optimizing, and the Dell PowerEdge line reflects these divergent needs. "The financial services industry has a latency concern for certain parts of their workloads," he says. "And they want to make sure their data is properly tiered because data can be the critical path through the system. So they are spending a lot of money. Across enterprise IT, they want one standardized system that can fit many of their workloads because they do not have the engineering capability of an HPC or hyperscale shop. So they want modularity, automation, and standards."

As for trends, there are a few that Gorakhpurwalla and his team are watching. For one thing, there are somewhere between 8 million and 12 million servers out there running Microsoft's Windows Server 2003, which will follow the Windows XP operating system into the bit bucket, in this case in September 2015. In most cases, companies will use the end of maintenance on Windows Server 2003 as an excuse to upgrade their systems, and the Haswell Xeon E5 launch this summer gives them the new hardware on which to port their applications to Windows Server 2012. But it is more than that.

"We know from our larger and more sophisticated customers that they want not just a change of iron, but a change of capability with that," says Gorakhpurwalla. "That means moving one step closer to converged or towards private cloud or hybrid capability. So I think this is opening up a big seam for a lot of our customers to get over to future-ready IT."

EnterpriseAI