In-Depth
Q&A: HP – Recasting Windows as a Giant-Killer
The new system is being promoted as an extensible platform capable of hosting for Windows Server 2003 Datacenter Edition, HP-UX, Linux, and even OpenVMS workloads
In November 2003, Hewlett-Packard Co. announced its 64-way Integrity SuperDome, the largest Windows server to date. At the time, HP took dead aim at the entrenched mainframe and RISC-Unix markets (other people’s Unix, that is), positioning the new system as an extensible platform capable of hosting Windows Server 2003 Datacenter Edition, HP-UX, Linux, and even OpenVMS workloads.
Since then, says Jeffrey Kyle, product marketing manager for Windows on Integrity, HP’s 64-way behemoth has sold well. While no customer has configured Integrity SuperDome to run Windows across all 64 processors, a surprising number are running Microsoft’s flagship operating system on as many as 32 processors.
What’s more, says Kyle, Windows Server 2003 Datacenter Edition packs a compelling value proposition, so much that existing mainframe and RISC-Unix customers are willing to overlook its relative immaturity.
What kind of demand is HP seeing for very large servers running Windows? Is there volume, so to speak, for these large Windows servers?
I think what we’re seeing is very good uptick, or strong interest from customers, and those are not in the spaces of necessarily moving a standard Windows file and print environment over to a big server, but in areas of business processing, especially business intelligence and data warehousing, where RISC-Unix is very strong today, and customers have a demand or a need to move to a Windows environment in that space.
Also, there’s demand from some existing [Windows] customers. A company that has grown up through a dot.com and survived the explosion and has used [Windows] for years, they have needs that need to be addressed in terms of growth. These are where the big systems are selling today, and are of great interest.
What are the largest Windows servers you’re selling today?
Windows Server 2003 DataCenter Edition in particular supports up to 64-processors today, and we have the benchmarks behind it to prove that it will support that.
How much memory are we talking about on a system of that size?
Today, one-half of a TB [512 GB]. With [Service Pack 1], 1 TB.
Are customers using the entire complement of processors in a single system image? I mean, are any of them running Windows 2003 [DataCenter Edition] across all 64 processors in a SuperDome system?
You’ll have seen from other vendors that have the big servers in this space … that there’s a lot of partitioning and consolidation. But the fact that we did come out with a 64-way system sort of pushed the envelope, so we do have customers interested in running 64-way systems, and we have systems in production with 32 processors.
That’s 32 processors in a single Windows image?
Yes.
What kinds of applications require large Windows systems like these?
Things like business intelligence, data warehousing, online-transaction processing—these are the typical applications that we see. Situations where [customers] need the 64-bit [memory] address space and the processing power that Itanium has.
Many of the vendors I’ve talked with have stressed that it’s not so much the aggregate size of the system but how well you make use of it—e.g., the extent to which you can dynamically partition it and allocate or de-allocate resources. Is that a view that HP endorses?
That is a position for the high-end server market in general, a classic RISC-Unix market … where you needed partitioning resource management and all of these other things that you have mentioned. For customers who are moving to Windows and exploring the TCO benefits of Windows, a lot of them are doing it because there’s a pent-up demand to [move away from RISC-Unix and] go to other systems, so we’re addressing that right now without necessarily the mature partitioning schema that is associated with Unix.
But you do support Windows partitioning of some kind on SuperDome?
Clearly, we support partitioning on our SuperDome systems and our midrange systems as well, and clearly customers are interested in the hard partitions that we provide. So I can partition that system and it can be electrically isolated from other systems, so I can take that system down and work on it without affecting the other systems. I think we’ll be working with other customers to understand how to utilize resources on an ongoing basis … as Windows matures, so we’ll be able to move to more-sophisticated partitioning schemas.
Even with these limitations, you’re seeing displacement of entrenched RISC-Unix and other systems?
Yes, that’s definitely a place where we are having success right now. [Windows is] especially of interest to customers [who] have mainframe systems, where the high cost of licensing and services takes up a lot of the budget on an ongoing basis. A system like SuperDome has single-instance scalability, but it also can be hard-partitioned for better availability. So you can take these workloads and [hard partition] them, so if one of them goes down, it won’t take the rest of the system with it.
Mainframes workloads and mainframe hardware are infinitely configurable, however. Does Windows provide anything that can really approximate that, especially for customers who are accustomed to it?
So far, with Windows itself, we have Windows System Resource Manager, which is a good first step to get into that. Currently, with the Windows operating system and the way applications run in Windows, Windows [System] Resource Manager is good, but we potentially look to Microsoft and third-parties to make better use of the system as the environment matures.
What capabilities does something like [Windows System Resource Manager] provide?
It can set a baseline [for utilization] in an environment, so if I’m running different applications, running those under Resource Manager will allow me to control that application if necessary and not let it get out of hand with memory usage. So certainly I am able to better manage what I have now with resource management and Windows System Resource Manager. But customers are able to do more than just set that baseline. They’re able to meter the usage of CPU and memory for supported applications, and they’re able to generate reports, [such as] “What am I doing?”, “What’s using resources in my systems?”, and so on, and it’s an education for those customers to see how Windows handles that. [Windows System Resource Manager] actually does pretty well.
So [for] customers who are used to more with RISC-Unix or mainframes, there’s the kind of expectation that these capabilities will get better as Windows matures?
That’s a pretty good assessment of Windows in general. [Windows System Resource Manager] doesn’t do things like I/O [management], which you get in a classic Unix environment, so the level of maturity is increasing, and if you start to put Windows on a server like SuperDome, you start to see maturity. I don’t think customers today by and large expect Windows to match Unix tit for tat, so to speak. They expect it to continue to improve.
So customers deploying the large Integrity servers with Windows are starting out with one set of expectaitons, and as their experiences progress, the opportunities they see to be able to deploy on the Windows servers grow.
Let’s talk about customers who want to run HP-UX and Windows together. Is it possible to mix and match these workloads in the same cabinet? Do customers often choose to do so?
For enterprise customers who are running multiple workloads and multiple operating systems in their data centers, they all have different requirements. So what we need to do, though [to support this], is separate those [operating environments] into the hard partitions I talked about earlier, but we do support that and … we can put Linux and OpenVMS into a hard partition as well.
Is that a proposition that’s successful with some HP-UX customers who are upgrading and also have lots of distributed Windows servers?
I wouldn’t say that we are out there trying to migrate our customers from HP-UX to Windows.
Yes, I understand that. But if they’re upgrading from, say, older N-class hardware, and they also have all of these Windows servers sitting around, why not put everything on the same box?
That is a proposition that we go to each of these customers with, and I have to say that many of them right now are doing that. Many customers have SQL Server databases in disparate workgroups or departments, and there are literally hundreds and hundreds of these workloads out there. So there’s a great proposition to bring these in and put them in one place.
One last question, that’s perhaps apropos for both you HP-UX and Windows customers. … Itanium is said to be a comparatively more difficult processor for which to program than [Intel’s] CISC and the different RISC architectures. How has HP responded to this issue, and what are you hearing from customers who are developing on Itanium?
I think it’s still part of the education process from companies such as HP, Intel, and Microsoft, and this year we’ve just concluded what was called “Route 64” from Microsoft, and it was a series of seminars worldwide targeted at ISVs to help them understand what it takes to migrate code from 32-bit to 64-bit and to build your 64-bit applications. HP and Intel have also gotten together to do a similar course … [called HP/Intel Developer Forums 2004], which are three-day classes for developers to attend to understand this.
So what you mentioned early on is about the difficulty, so I think that the education that still needs to occur is that the difference between an Itanium processor and an x86 in terms of application development is that I have a lot more opportunities with Itanium. I can build an application and compile it for x86, and I can do the exact same thing with Itanium, but I can go in and do more, I can go in and write the assembly language, and tune, and do the assembly language in Itanium.
With SP1 for Windows Server 2003, when that comes out, we’ll be providing the .NET Framework, the Common Language Runtime, native in the operating system, and then when Visual Studio 2005 releases, [in the] first half of 2005, it will include native development for Itanium, as well as N64, and at that point, I can write managed code for Windows.