The Server Room Blog

Nehalem-EX Brings New Economics to Scalable Systems

michael.j.demshki@intel.com — Thu, 11 Jun 2009 16:10:42 GMT

The debate on how to best increase system capacity to accommodate growing applications has raged on for years; “scale up” with more CPU, memory, and I/O, or “scale out” with loosely connected systems. Scaling out by adding networked systems to increase capacity has been a good economical solution for many IT managers because it allows them to grow by using less expensive, industry standard building blocks. However, there are some notable exceptions to this line of thought. One is that the class of applications that require shared memory and large database support are much better suited to run on a single, expandable system that scales up. These are typically transaction processing, business intelligence and ERP solutions. Until now, IT managers running applications that require scale-up systems larger than 4 or 8 CPUs have had limited platform choices and most were proprietary and expensive RISC-based servers.

The other problem with the scale out approach is the people, facilities, software and overhead costs and complexity of managing very large numbers of servers, which can grow to a point where the costs outweigh the performance and system cost benefits. The industry solution to achieving better ROI has been to consolidate multiple scale-out servers onto single industry standard scale-up servers with virtualization solutions. This is a good solution, but is limited by the number of application loads the IT manager feels comfortable placing on a single server, given the need to maintain peak performance and availability for each application.

Well, it looks like the scale-up, scale-out debate is about to take another turn. In the server product update Intel gave on May 26th, they talked about new levels of system scalability and choice supported by the upcoming Nehalem-EX processor. This processor will support systems that scale up to 8 sockets natively (shared memory, without any additional silicon), and up to 16 sockets and higher with node controllers from system manufactures that allow single systems to share memory beyond 8 sockets. So far there are over 15 different designs from 8 OEMs that offer 8 socket or higher scalability. But of course, for the class of application where scaling is important, socket count doesn’t tell the whole story of what’s needed for scalable performance. Thread support, key for transaction processing and virtualization, scales at the rate of 16 threads per socket with 8 cores and Hyper Threading (2 threads per core). That would be 128 threads for an 8-socket system, and 256 threads for 16 sockets. And in order to keep those threads fed with data close to the CPU, each processor supports up to 24 MB of shared cache (1.5X current generation Xeon), and an impressive 16 memory slots per socket or 128 DIMMs on an 8-socket system. In addition, the Scalable Memory Interconnect gives these systems 9 times the memory bandwidth of today’s top Xeon processor. Finally, four QuickPath interconnect links per socket allow for high-bandwidth sharing of data across the system.

So the net of it is that the industry is going to see a broad selection of highly scalable, next-generation servers that significantly extend the economic advantage of industry standard scale-up solutions for business-critical, large database, and high-end virtualization/consolidation deployments. I would expect these systems to give IT managers a very cost-effective alternative to the much more expensive and proprietary RISC-based servers they use today.

What are your thoughts? Mike

When Xeon 5500 series “Nehalem” servers are just not enough…

bryce.olson@intel.com — Wed, 22 Apr 2009 01:00:42 GMT

Sure, Intel® Xeon® 5500 Series Processors represent a quantum leap forward in terms of both performance and energy efficiency. That has been proven in a number of test results and reviews. But for your back-end data demanding enterprise app deployments, large scale server consolidation or virtualization of business critical applications, Intel® Xeon 7400 series processors offer outstanding performance and performance per watt in 4-socket servers. So, which platform do you choose, especially when this decision is likely going to be the key determining factor for capital savings, efficiency and TCO for your datacenter infrastructure? Well, you’re read a lot about Xeon 5500 series Nehalem servers over the last few weeks. Let me share with you some reasons to consider a Xeon 7400 series 4-socket server when you are presented with the choice between Intel’s two best of breed products for virtualization.

4 Socket and above servers (Xeon 7400) are purpose built – just like a large truck: They’re purpose built for your most data demanding enterprise applications like database and ERP, and for large scale server consolidation using virtualization. Large Trucks are also purpose built. They’re purpose built for hauling large loads over long distances. Now, you don’t buy a large truck to commute to work in. You also don’t take your everyday commuter and attempt to haul large loads with it, because if you did you would be significantly undersized (you’ve all seen those cars on the road with rear tires about ready to pop under the weight of a palette of heavy goods tied on top).

More Resources Matter for 4 Socket MP Workloads:
The apps/workloads listed above benefit from the expanded feature set associated with 4 Socket Xeon 7400 based servers: more processors (4 vs. 2), more cores (24 vs. 8), more memory (32 dimms vs. 18 dimms), more I/O capacity (7 slots vs. 4) and larger cache (16MB vs. 8MB). These features and what they enable are why MP Server buying patterns have remained stable with IT for the last 5 years and will continue to be stable for the foreseeable future according to IDC.

But in today’s economy there may be MP customers out there that will want to push the envelope and attempt to deploy lesser expensive 2S systems for traditional 4S solutions. Would doing so pencil out from a TCO perspective? Let’s take a look at two Virtualization usage examples and find out.

Large Scale Server Consolidation: Where almost 2x the memory matters.

In this scenario, IT Manager is dealing with numerous corporate acquisitions across the country prior to the economic downturn, with servers that now need to be consolidated to cut costs quickly. Goal is to convert 1000 older underutilized 2S servers. He (she) converts these to 1000 VMs and transfers them electronically to the central Data Center. He determines that these infrastructure apps when consolidated generally run into memory constraints before they run into processor constraints, so for his candidate solutions he compares a 4 Socket Server with Xeon X7460 processors vs. a new 2 Socket server with Xeon X5570 processors. He fully loads both systems with 4GB dimms (128GB on 4S vs. 72GB on 2S), and assigns 4GBs memory for each VM deployed (enabling 32VMs per server resulting in 31 new 4S servers vs. 18 VMs per server resulting in 56 new 2S Servers.)

Now, he only propagates the 4S Solution with 2 Xeon 7400 Processors, which allows the IT manager to still use all 128GB of memory on the 4S Servers while paying lower VMWare licensing costs. Price these systems out on Dell, HP, IBM’s or Sun’s website, and the Xeon X7460 servers will be in the $15k-$20k range vs. the Xeon X5570 based servers will be in the $10k-$12k range (i.e. roughly 1.5x higher for 4S vs. 2S server). Add VMWare license costs, power/cooling, LAN/SAN cabling, and system maintenance costs and you’ll see the 4S solutions offer a lower cost per VM.

Virtualizing Business Critical Workloads: Where 3x the Processor Cores matter.

In the previous example, we were looking to maximize consolidation ratios. In this example, we’re looking to achieve predictable high performance for a business critical app. Solutions like ERP that are put into a virtualized environment perform best when run without oversubscription, where you set the same number of virtual CPUs to equal the number of physical cores available on the platform. This helps deliver relatively more predictable performance for all VMs and is the way that IT@Intel intends to deploy ERP in a virtualized environment as they begin to test this moving forward (read more about this in the new whitepaper). In this example, we’ll convert ~100 non-production ERP instances (i.e. the instances used for QA, Dev, and Production break fix). We’ll assign 2 virtual CPUs and 8GB memory for each instance. The four-socket Xeon 7400 processor based systems (with 96GB memory) will have a total of 24 cores and will have a list price of about $25k. This allows us to run 12 Virtual Machines without oversubscription on the MP Servers and enables 100 ERP instances to be consolidated down to about 8 MP (4 Socket) servers. Since the Xeon 5500 based Servers just have 8-cores, the IT manager decides to avoid oversubscription and deploys 4 virtual machines – consolidating down to 25 DP (2 Socket) servers with 32GB Memory and a list price of about $8k per server. Include the costs of the hardware, VMware ESX license costs, power/cooling, cabling, and Server maintenance – the MP (4 Socket) solution here would also offer a lower cost/vm than the Xeon 5500 based DP (2 Socket) solution due to having 3x the processor cores on 4 Socket.

When you are deploying your most data demanding enterprise applications and implementing large scale server consolidation, Xeon 7400 based servers represent a very intelligent choice.

Let me know what you think.

bryce

Should I refresh my servers or just upgrade my CPUs ?

alan.priestley@intel.com — Wed, 08 Apr 2009 23:31:01 GMT

There has been lots of discussion recently about whether its better to replace or upgrade existing CPUs in your installed base of servers rather than purchase new servers. I wanted to share some thoughts with you that might clarify why a new server purchase is the better option for most IT departments.

Here are some of the challenges that an IT department must face when considering replacing CPUs rather than buying new servers.

Does the existing system support the new CPU – most CPUs require specific BIOS versions, is there a BIOS update available for the server that supports the new CPU ? Also the server motherboard may not have been tested by the OEM with the new CPU.
Has the software stack you are running on the server been validated on the new CPU.
Replacing a CPU is a non-trivial exercise – it takes time and you run the risk of damaging a working server
- the server must be shut down and dismantled to access the existing CPU
- the existing CPU/heatsink combo must be removed. The heatsinks used by OEMs in branded servers are specifically designed for the server in which they are used. These heatsinks typically have significant mass so they are usually very firmly attached to the server chassis to prevent damage due to shock and vibration whilst in transit and in use.
- The existing heatsink must be removed from the current CPU, which may not be easy if the system has been in use for some time the thermal bonding may have hardened permanently attaching the heatsink to the CPU.
- The heatsink must be attached to the new CPU – with the appropriate thermal bonding.
- The CPU/heatsink combo must then be correctly re-installed in the system and the system re-assembled.
It is also necessary to take into account that removing/changing a CPU may also void or otherwise affect the system warranty.

It is conceivable that some IT folk may want to consider this approach but the risks associated with undertaking this operation are very high and many IT departments will take the approach of not touching working systems. If you are still not convinced its also worth considering -

Replacing the CPU in an old server may not significantly improve its energy efficiency. The latest generation server designs not only use latest CPUs but they incorporate many new features that improve the overall energy efficiency of the complete platform – making them a much better proposition when looking to reduce overall data centre utility bills and OpEx costs.
Upgrading the CPUs in an old server may expose other limitations of the server in terms of memory and I/O, this could result in having to upgrade many other parts of the server resulting in an overall higher cost than replacing the server with a new purpose designed solution

So, as far as I can see very few IT departments are going to seriously consider replacing CPUs in their existing installed base and will look instead to deploy latest generation high performance energy efficient server designs – i.e. servers based on Xeon 5500 or Xeon 7400

What’s your opinion – are you prepared to attempt to upgrade your CPUs or will you refresh the complete server system to get the latest technology for all elements of the server platform

Inverting Moore’s Law Means Big Cost-Savings for IT in 2009

bryce.olson@intel.com — Fri, 06 Mar 2009 20:05:41 GMT

In 1965, Intel co-founder Gordon Moore made a prediction, popularly known as Moore's Law, stating that the number of transistors on a chip will double about every two years. Intel has kept that pace for nearly 40 years. For IT, this translates into a roadmap that enables IT to buy new servers that cost roughly the same as the previous server but performs so much better. Compare Intel’s 4 Socket MP server performance introduced in 2006 (Intel Xeon processor 7000 series) to today’s server introduced in 2008 (Intel Xeon 7400): 3x more performance throughput as measured by SPECint*_rate_base 2000*, 2.4x more ERP users as measured by SAP-SD* and 2x more database transactions as measured by TPC-C*.

Now, introduce a global economic downturn into the mix and suddenly IT is forced to cut costs and projects (i.e. delay or cancel upgrades and non-revenue generating projects). New articles start popping up from magazines like the Economist that take Moore’s Law and propose flipping it on it’s ear: instead of products providing more performance at roughly the same price, provide products that offer the same performance as IT is already experiencing, but now at a lower price. Call it “inverting Moore’s law” where IT takes the dividend it provides in dollars vs. extra performance.

So here’s something to think about: You can also “invert” Moore’s Law by making new targeted IT investments today that offer attractive payback scenarios tomorrow - giving you similar performance but at a much lower cost. With mortgage rates dropping, you may have already benefited from a rapid payback in your personal life (i.e. I recently refinanced a house down from 7% to 5.25% 30-year fixed rate that I had continuously made additional principle payments for. The ~$5k up front investment (i.e. closing costs) will be “paid back” to me after 5 months due to monthly mortgage payment savings.

Here is a server refresh example that explains how you can also get an attractive payback for your IT department.

Oracle Database Refresh: Let’s next look at a hypothetical example of an IT department running current Oracle Database Enterprise Edition on 12 servers purchased in early 2006 (dual-core Intel Xeon 7041 based servers introduced in 2005) and assess the total cost of ownership difference in moving to new servers. We’ll assume the IT manager is paying per processor licensing fees for Oracle Database. We’ll compare the old server equipment to new 4-core Xeon 7440 based servers that offer up to ~3x more database performance (Xeon processor 7400 Series come in flavors of 6-core and 4-core versions). This should enable consolidation ratios of 3:1, enabling the IT manager to reduce from 12 servers to 4 new servers.

First the new investment: 4 New Xeon 7400 based servers at roughly $20k each = $80k. Add another $5k for Network, Server Maintenance and Install Costs. Remove ~$2k in tax implications associated with the expense in year 0. Total investment ~$83k.

Next, let’s look at the savings: The IT Manager is paying $41.8k yearly on Oracle maintenance/support costs x 12 dual-core MP servers today, that is $501k. The 4 new quad-core servers will have larger Oracle database maintenance/support costs because of the core count ($83k x 4 servers = $334k) but this will still result in $167k SW savings each year (difference between $501k and $334k) which my calculations show about $669k savings over 4 years. Moving from 12 to 4 servers also reduces about $72k in network, server maintenance, and utility (power/cooling) costs over 4 years as well. In addition to all of these costs savings over 4 years, my calculations show that the original investment of ~$83k has a payback of 9 months.

Targeted IT investments today can offer attractive payback scenarios and cost savings tomorrow - giving you similar performance but at a much lower cost. Let me know what you think?

Have you got the "VIBE"?

william.h.lea@intel.com — Mon, 29 Sep 2008 22:00:03 GMT

Check out this video with Robert Harley of Virtual Blocks and Chris Parisi with HP Canada talking about a new virtualization solution for SMB's.

Dunnington Delivers Energy Efficient Performance

david.e.jenkins@intel.com — Wed, 24 Sep 2008 12:18:47 GMT

Good news for the enterprise - the latest "Tick" of Intel's "Tick Tock Model" has made its way to the high-end 4 socket segment and the energy efficiency improvements are sure to make an IT manager smile.

With the launch of Intel® Xeon® Processor 7400 series, the entire Intel® Xeon Processor product line is now using 45nm process technology, hafnium based hi-k dielectrics, metal gates and enhanced Intel Core Microarchitecture. The results are just what you have come to expect - improved energy efficient performance because of higher performance delivered by more cores and processor architecture improvements using faster, lower leakagetransistors.

What does this really mean to you?

Do you need better performance? How does up to a 50% improvement over previous generation Intel® Xeon® 7300 processors sound?
Is power a concern? A server configured with Intel® Xeon® 7400 Processors consumes ~equal or less power than the previous processor generation.
Combined, the performance and power improvements deliver up to a 54% improvement in energy efficiency.
Given the breadth of Intel® Xeon® 7400 Processor choices - from 6 core 2.66GHz (130W TDP) processors down to 2.13GHz (50W) 4-Core to the 2.13GHz (65W) 6-Core that is the lowest power per-core processor on the market, you can choose the right processor to deliver the balance of performance and power that meets your compute needs.

In summary, with the Intel® Xeon® 7400 Processors, you can deploy the same number of servers in your data center while increasing your performance capacity or deploy fewer servers to complete the same amount of work while reducing power consumption. Using the best energy efficiency servers is a great first step toward increasing the efficiency and performance of your datacenter - look for a follow on blog later this week from Dave Hill to talk about other actions that you can take to reduce your power consumption and carbon footprint too.

“VMWorld08: Intel Keynote-Collaboration for Leading Virtualization Deployments”

william.h.lea@intel.com — Thu, 18 Sep 2008 18:28:49 GMT

More news from VMWorld 2008, Las Vegas. Doug Fisher, Intel V.P. gave a keynote during the VMWorld conference. One of the more interesting elements brought Steve Herrod, Sr. V.P. and CTO of VMware on stage to talk about how Intel and VMware are collaborating to deliver leading Virtualization Deployments. Click on the video to see what they have to say.....