Well, it is
Thursday morning, and the last day of Tech @ Work. I'm catching one last
session before heading to the airport. While I often blog about HP-UX, this
time I'm trying to learn a little more about Windows, especially for mission
Grizaud, the speaker, started out covering some of the trends in the industry:
things like pressure to reduce underutilized servers, reducing physical servers
using virtualization, and things like that.
Laurence talked about the needs to be scalable,
reliable, and operationally efficient. While scale out is one way to increasing
resources, she focused on scale up solutions. Windows runs on everything from a 2 socket server to a
Superdome with 64 sockets and 128 cores and 256 threads. Windows 2008 R2 now
supports up to 128 cores, up to 2 Tb of memory, and up to 192 I/O slots that
are available on the Superdome. This provides a stable platform for large OLTP,
Business Intelligence, and Data warehouse workloads, for server consolidation
using VMware, Hyper-V, or database instance stacking, or for I/O or memory
constrained applications that require the added scalability.
At the end of
the day, the scaling requirements of your application will usually determine
whether you use a scale up or scale out model. For instance, SQL 2008 R2 will
be able to balance its workload across all 256 logical processors (threads) on
a current Superdome. HP offers servers that can work in either scenario.
If you are going
to add a whole bunch of workloads to a single server, the impact of an outage
is much more serious since it impacts more workloads. That usually requires a
more reliable and resilient infrastructure. A more reliable system means less
problems. A more resilient server means that if there is a problem, the server
doesn't fail and bring down the workloads. This extends to extended distance
clusters to provide availability even if a server, or datacenter, goes down.
is a requirement for operational efficiency. Most customers have Window servers
running somewhere in their environment. If you are looking for a scale up,
mission critical Windows server, it should use the same tools and processes as
you already use on smaller systems. HP includes the HP Insight Foundation, and
offers the HP Insight Control and HP Insight Dynamics across all of our servers
that offer Windows support.
critical Windows servers - offering scalability, resiliency, and operational
efficiency. Do you use scale up Windows servers? If so, what applications do
you run, and what hardware do you utilize.
Well, this is
just about it for me at Tech @ Work this year. It has been a great time with
the new product introductions, meeting customers and colleagues, and listening
to some great speakers. I'm on the road for a few weeks in the APJ region, so
blog updates might be a little more sparse. Hopefully you have found the team
coverage of Tech @ Work interesting.
I've read some feedback about this blog that it seems to focus on HP-UX 11i and high end servers. While HP-UX 11i, OpenVMS, HP Integrity Non-Stop, and high end servers in general are definitely designed with mission critical workloads in mind, they aren't the only mission critical systems that are available, even within HP. Today, we'll look at a couple of other options.
First, HP has announced its availability of Windows Server 2008 R2 (PDF - http://h20341.www2.hp.com/integrity/downloads/Announcement_page_for_Windows_2008_R2_for_BCS_Scal.pdf). One of the key upgrades is support for 256 Logical Processors, allowing Windows to scale to the full 64 sockets, 128 cores, and 256 threads of a HP Superdome. There is even a demo available at http://h30431.www3.hp.com/?fr_story=108083959c3c5b8d6a20c2e7ff222e4eefa0d9e2&rf=bm.
Why is this important? While I spend a lot of time dealing with HP-UX, I have met a number of customers who have standardized on MS SQL server as their database of choice for their business. I've spoken to customers who run credit card businesses, all on Windows and SQL server. I've heard about tax authorities running on Windows. I've heard of large SAP instances running on Windows at a brewing company. I know of plastic manufacturing companies who rely on Windows on HP Integrity servers. These organizations and many more, likely could take advantage of the additional scaling, since they run very large workloads on these systems. Oh, and scaling isn't just about throwing more processors at a workload. A system needs to balance processor, memory, and I/O capabilities, and HP Integrity servers do that.
However, scaling isn't the only feature, or even a requirement, for all mission critical systems. One statistic that I've heard, although I don't remember where, is that Windows crashes approximately 1/3 of the time on Integrity versus a standard x86 server. Why? HP Integrity servers, at the high end and the entry level, have additional hardware reliability features, such as double chip spare technology so that it can handle double bit memory errors, something that will make many other servers reboot.
But what if you want to run mission critical Windows, get some reasonable scaling, but don't want the extra reliability of HP Integrity servers? Naturally, we have HP ProLiant servers, including the DL 785 G6, which now scales to 48 cores. I've spoken with a number of sales reps about this system, and its predecessor, the DL 785 G5. I've consistently heard two things about workloads on this type of server: database and consolidation. While a big box for database use is a pretty straightforward use case, using a big box for consolidation using virtual machines takes a little more thought.
Why? Well, one of the benefits of server consolidation and virtualization is the reduction in overhead. If you have one workload, you need to size it for peak demand, even if the average demand is much lower (let's say a peak of 8 CPUs, but an average of 3 CPUs). If you had two standalone servers with this workload size, you would need 16 CPUs. But, if these workloads don't peak at the same time, you might be able to get away with a single server with, say, 12 CPUs, a 25% savings. Oh, and HP Capacity Advisor, part of both the HP Virtual Server Environment (http://h71028.www7.hp.com/enterprise/cache/258348-0-0-225-121.html) for HP Integrity servers and HP Insight Dynamics - VSE (http://h18004.www1.hp.com/products/servers/management/insightdynamics.html) for HP ProLiant and HP Bladesystem x86 servers, and show you exactly how much capacity is actually required.
But, the big savings isn't from sharing two workloads. As you increase the numbers of workloads, the chances of all the workloads being busy at the same time decrease. That means that you need fewer CPUs, memory, and I/O to handle all the simultaneous workloads. When used in this manner, larger servers can run a lot more workloads than smaller servers (as a general rule of thumb), and operate at higher levels of utilization, and require less hardware to get the same amount of work done. Whether it is a HP ProLiant DL785 running or a HP Superdome, the idea is the same. And with the increase in Windows scalability, both cases work well.
Do you use Windows in a mission critical environment? If so, what workload are mission critical for you? Have any interesting stories around this? If so, let me know.