Energy Efficiency Another POWER7 Benefit

More and more companies are becoming sensitive to their energy costs. Though the idea of going green inspires some jokes (if the walls in my datacenter are painted green, does that make me part of the green computing movement?), the reality is that energy costs are high and getting higher.

Rising cost isn't even the whole story. I've seen datacenters that were running at maximum capacity -- they literally couldn't run any more circuits. Obviously consolidation should be a serious consideration in these cases, but even bigger steps are being taken.

A customer recently told me over lunch that some of his company's largest expenditures are energy costs. In response, they're planning to get off of the grid completely. The goal is to use solar and wind power, and to generate energy internally by burning biomass. I've heard other companies are looking to geothermal for heating and cooling.

An interesting and, perhaps, overlooked aspect of IBM's recent POWER7 announcement is the energy savings potential of these systems.

From IBM:

"The new systems can deliver four times the performance and four times the virtualization capability for the same price, and are three to four times more energy efficient...

"IBM Power 750 Express, an Energy Star qualified business server for mid-market clients [offers] four times the processing capacity of its predecessor, the IBM Power 550 Express...

"POWER7 technology features 'Intelligent Threads' that can dynamically vary based on workload demand. With more threads, POWER7 can deliver more total capacity as more tasks are accomplished in parallel, such as monitoring the energy usage of millions of households by the minute in a smart grid. With fewer threads, those workloads that need very fast individual processing -- such as real-time analytics or database transactions -- can get the performance they need for maximum benefit. Intelligent Threads work on all POWER7 processors...

"IBM's POWER7 systems are designed to make dramatically better use of energy. Unique Intelligent Energy technology allows customers to power on and off various parts of the system or to dynamically increase or decrease processor clock speeds based on thermal conditions and system utilization, on a single server or across a pool of multiple servers. POWER7 energy management technologies are integrated from its processor, to firmware, PowerVM virtualization, operating system support, and up to IBM Active Energy Manager software, included in the new IBM Systems Director Standard and
Enterprise Editions. As a result, the system dynamically balances between energy usage and performance and systems utilization based on policy. The result is improved performance per watt -- more than two-times better than similar Intel x86-based systems, four times better than Sun SPARC servers and eight times better than similar HP Itanium-based servers."

As we dig into the details of the POWER7 announcement and get excited about the performance gains we can expect, don’t overlook the greater energy efficiency. It's one of the biggest benefits we'll see as we upgrade our hardware.

More on Migrating to Virtual Storage

The now published IBM Redbook, "PowerVM Migration from Physical to Virtual Storage," contains a wealth of valuable information. Here are a few more highlights.

* Choosing a SAN disk -- The authors of the Redbook detail the various methods for determining which SAN disk to use when virtualizing: Physical Volume Identifier (PVID), IEEE Volume Identifiers and Unique Device Identifiers (UDID).

* The new chkdev command -- Lots of material here. As the authors state:

"As of Virtual I/O Server Fix Pack 22, a new command has been introduced to assist with the identification of disks and their capabilities. The Virtual I/O Server chkdev command is capable of displaying the same values as mentioned previously (IEEE, UDID, and PVID) but provides some additional information:

#chkdev
NAME:                hdisk0
IDENTIFIER:          2F1135000C50006D9C1F30BST9146802SS08IBM-ESXSsas
PHYS2VIRT_CAPABLE:   YES
VIRT2NPIV_CAPABLE:   NA
VIRT2PHYS_CAPABLE:   NA

* What can be virtualized? Again, I'll let the authors tell you:

"PHYS2VIRT_CAPABLE: This disk may be virtualized to a logical partition. Once this is performed, this field will change to a value of NA if the mapping is successful. A value of NO indicates this volume may not be virtualized.

"VIRT2NPIV_CAPABLE: If the disk is capable of moving from a virtual SCSI environment to an N_Port ID Virtualization (NPIV) environment, this field will be set to YES, otherwise it will be set to NO. A value of NA means that this disk has already been moved and the Virtual Target Device or VTD as it is abbreviated in the command output, will indicate the mapping.

"VIRT2PHYS_CAPABLE: If the device is capable of moving from a virtual environment to a physical environment and is currently mapped to a VTD then the value here will be YES. A value of NA means the disk is not in use by a VTD, while a value of NO means the disk is not capable of such a move. For further information, please refer to the Virtual I/O Server chkdev manual page."

* It figures -- The Redbook also includes some nice charts (see the Figures section in the table of contents) that help make concepts clear. For example, Figure 2-1 shows the relationship between physical SCSI disk and the target SCSI device on the Virtual I/O Server. Figure 2-2 shows the relationship between physical SCSI disk and the virtual SCSI devices on a client partition.

* NPIV -- As we reduce our reliance on physical adapters, and with more customers adopting NPIV, it's nice to see this topic covered:

"N_Port ID virtualization (NPIV) is a technology that allows multiple logical partitions to access independent physical storage through the same physical Fibre Channel adapter. Each partition is identified by a pair of unique worldwide port names enabling you to connect each partition to independent physical storage on a SAN. Unlike virtual SCSI, only the client partitions see the disk. The
Virtual I/O Server acts only as a pass-through managing the data transfer through the POWER Hypervisor."

Finally, there's this interesting note on page 29:

"A POWER Hypervisor has a limit of 32,000 pairs of WWPNs. If you run out of WWPNs, you must obtain an activation code for an additional set of 32,000 pairs.

"Each time you configure a virtual Fibre Channel adapter, whether dynamically or by adding to a partition profile, the HMC obtains a new, non-reusable, pair of WWPNs from the POWER Hypervisor. Therefore, the correct procedure for dynamically allocating a virtual Fibre Channel adapter to an active partition that must keep the configuration across a partition shutdown is to first dynamically allocate the adapter to the partition and then to use the HMC Save Current Configuration feature to save the configuration to a new profile. This new profile then must be used to start the partition after a shutdown. This will ensure that the WWPNs that were allocated during the dynamic operation will be the same ones in the profile. If instead, you dynamically add an adapter and then add an adapter to the
partition profile, the partition will come up with a different pair of WWPNs after a partition shutdown and access to the storage will be lost."

This summary -- and this, and this -- barely scratch the surface when it comes to summarizing the valuable information in this publication. So take the time to read this Redbook.

Also, some interesting response to the POWER7 announcement:

* POWER7 facts and features

* "POWER7 has the most remarkable on-chip cache hardware of any processor on the market."

* Smarter systems for a smarter planet

* POWER7 does not validate Sun's CMT SPARC Processor Architecture

* Active Memory Expansion Performance

* Active Memory Expansion Overview and Usage

* Power your planet with Power Systems

Active Memory Sharing Takes the Guesswork out of Partitioning

Like I wrote in AIXchange a few weeks ago, I was minding my own business when someone tweeted a link to a new IBM developerWorks article by Chris Gibson, "Configuring Active Memory Sharing: From a customer's experience."

I'd seen the movies that Nigel Griffiths had done based on this topic:

http://download.boulder.ibm.com/ibmdl/pub/systems/power/community/aix/AIX_Movies/AMS_RegularPaging.wmv

http://download.boulder.ibm.com/ibmdl/pub/systems/power/community/aix/AIX_Movies/AMS_Concepts.wmv

http://download.boulder.ibm.com/ibmdl/pub/systems/power/community/aix/AIX_Movies/AMS_Setup2.wmv

http://download.boulder.ibm.com/ibmdl/pub/systems/power/community/aix/AIX_Movies/AMS_Monitor2.wmv

I'd also read the IBM Redpaper, "PowerVM Virtualization Active Memory Sharing."

Chris' article, though, is yet another great source of information about this additional functionality. PowerVM Active Memory Sharing allows you to virtualize memory on your frame. You can overcommit physical memory and allow the memory to flow between logical partitions and paging space.

Virtualized memory operations aren't as instantaneous as virtualizing your CPU. While your micro-partitioned CPU makes changes on a millisecond basis, memory sharing uses paging space, so it's obviously orders of magnitudes slower as it reads and writes to disk. So if you have several logical partitions that are all starved for real memory at the same time, this isn't for you. However, it is great if your LPARs don't simultaneously demand memory. When you create an LPAR and allocate some
amount of memory, it's possible, even likely, to guess wrong. Allocate too much memory and it goes to waste. Allocate too little and you might run into performance issues. With Active Memory Sharing however, the system determines which LPAR actually needs the memory. This makes more efficient use of your resources.

From the Redpaper:

"Active Memory Sharing allows overcommitment of memory resources. Since logical memory is mapped to physical memory depending on logical partitions' memory demand, the sum of all logical partitions' logical memory can exceed the shared memory pool's size. Each logical partition is allowed to use all assigned logical memory. When the cumulative usage of physical memory reaches the pool's size, the hypervisor can transparently steal memory from a shared memory partition and assign it to another shared memory partition. If the removed memory page contains data, it is stored on a paging device and the memory page content is cleared before it is assigned to another logical partition. If the newly assigned page contained data, it is restored from the disk device. Since paging disk activity has a cost in terms of logical memory access time, the hypervisor keeps track of memory usage to steal memory that will likely not be used in the near future. The shared memory partition's operating system cooperates with the hypervisor by providing hints about page usage and by freeing memory pages to limit hypervisor paging."

Read Chris’ article and the Redpaper, and watch the movies. Think about the different ways that this functionality can be useful in your environment. Look at the list of requirements.

Again, from the Redpaper:

"In order to use the Active Memory Sharing feature of IBM PowerVM, the following are the minimum requirements:

"An IBM Power System server based on the POWER6 processor
Enterprise PowerVM activation
Firmware level 340_075
HMC version 7.3.4 service pack 2 (V7R3.4.0M2) for HMC managed systems
Virtual I/O Server Version 2.1.0.1-FP21 for both HMC and IVM managed systems
AIX 6.1 TL 3

"Shared memory partitions can be created on a system as soon as the shared memory pool has been defined. In order to be defined as a shared memory partition, the logical partition must meet the following requirements:

• Use shared processors.
• Use virtual I/O, including: virtual Ethernet adapters, virtual SCSI adapters, virtual Fibre Channel adapters, virtual serial adapters

The operating system running in the logical partition can be either AIX, IBM i, or Linux."

Start planning now for any necessary upgrades so that you can take advantage of this new way to manage your machines.

IBM Hardware Fits in the Big Picture

I've worked for numerous companies that were IBM customers. For several years I worked for IBM in Boulder, Colo. I currently work for an IBM business partner. Through all this time, I've been happily using IBM racks with IBM servers. Sure, there were other servers running other operating systems that used different racks in the various computer rooms, but the IBM servers all lived in IBM racks.

I was recently on a raised floor at a server co-location company and watched IBM customer engineers move a POWER5 Model 570 from an IBM rack to a non-IBM rack. To install the server, they had to carefully measure its required depth. To make the flex cable on the front of the machine fit into the rack, they had to modify the side of the rack due to some protruding metal on the side doors that was interfering with the flex cable. Obviously, the whole operation took longer than a regular rack install into a standard IBM rack would have taken.

While they got the server to work in the non-IBM rack, it wasn't an ideal situation. The power distribution units in the back of the rack were rotated compared to the IBM rack, and the plugs were oriented so that when the servers were plugged in, power supplies and PCI cards couldn't be swapped without removing the entire PDU first.

So why go to this trouble? These non-IBM racks are taller and have a smaller footprint than the previously used IBM racks. When you're paying by the square foot, you want to squeeze in as much equipment as you can. By using extra space at the top of the rack, the co-locator could fit more equipment into its floor space. I can certainly understand that with several taller, narrower racks, you can wedge more racks into a row of hardware, and that can translate into more revenue for the operators of the shared environment. (This might not matter so much in a customer-owned data center, but in this shared environment, it's critical.)

One of the facility managers said that IBM just doesn't get co-location. He argues that every other vendor can make its equipment fit into these racks, why can't IBM?

My take is that IBM engineering designs IBM racks with IBM hardware in mind. They're rugged. They're built to fit without modification, and they're built so that parts can be easily hot-swapped in and out of the machines. If IBM racks don't fit nicely into your cookie cutter server room that was designed with non-IBM servers and racks in mind, if they're wider than other enclosures, then maybe it's time to rethink the design of the room. The problem as I see it is that many customers think that everything is the size of an x86 machine. However, they should plan for equipment of other sizes. It's not necessary to redesign the computer room--but maybe designating an aisle for nonstardard-sized racks would give them some flexibility.

The thing is, the greater serviceability of IBM solutions is well worth the sacrifice of some floor space. IBM engineering designed and built its racks with IBM servers in mind. Using PowerVM and IBM Power Systems servers, you can consolidate multiple large workloads onto a much smaller footprint. And isn't creating the smallest footprint the goal here? Yes, IBM racks might take up an extra square foot or two compared to non-IBM racks, but this should be easily offset by an overall reduction in server hardware.

I believe IBM keenly understands co-location. They've given me hardware that I can easily carve into
multiple LPARs per physical frame, which allows me to consolidate workloads. The technology is very forgiving and flexible--as workloads grow and change, it's very easy to use a dynamic LPAR operation and make the adjustment.

Ultimately of course, the customer is always right. So if you need to make your server fit into a smaller footprint on your raised floor, you can and should do so. But first, reconsider what you're giving up by making this change. Saving physical space doesn't always equate to saving money.

A Look at Recent IBM Announcements

IBM made some new announcements last week. Take the time to get familiar with this information.

From the overview:

"IBM announced new Power Systems blades and servers, virtualization software and POWER6 technology, all designed to help companies build, manage and maintain a 21st century dynamic infrastructure."

Highlights include:

• New 4-core JS23 and 8-core JS43 blades using 4.2 GHz POWER6+ processors.
• Virtual tape support for IBM i and BladeCenter implementations. One thing that customers wanted was a way to natively back up their IBM i images directly to tape drive -- this announcement will help address that issue. Customers will be able to use a LTO4 SAS-attached single tape drive model TS2240. Once they upgrade VIOS, firmware on the blade and PTFs for IBM i (in the May 22 timeframe), they'll be able to use the tape drive as if it were natively attached.
• Power 520 Express and 550 Express servers with POWER6+ processors. The 520 has 2- and 4-core 4.7 GHz models, and the 550 will have 2-, 4-, 6- or 8-core 5.0 GHz models.

According to the overview, "The POWER6+ processor has also been shipping in the 3.6 GHz model of the Power 560, the 4.4 and 5.0 GHz models of the 16-core Power 570 and the 4.2 GHz model of the 32-core Power 570 since they were announced last October."

In addition, IBM has announced 69 GB solid state drives (SSD) to be used with Power Systems. I hear that you should expect performance increases when you put your "hot" data onto SSD drives versus
traditional drives. Using SSD allows you to use fewer drives (removing mechanical seeks and waits frees you from worrying about spindle count). This reduces power consumption and takes up less physical space.

Another thing I hear is that, to get to the most out of the SSD drives, you want many random reads and a low percentage of sequential/predictable reads, and you want a higher percentage of reads than writes.

More from the overview:

"PowerVM, IBM's virtualization solution, is now enabled for the first time with Active Memory Sharing, an advanced technology that allows memory to automatically flow from one logical partition to another for increased utilization and flexibility of memory usage." IBM published a Redpaper titled "PowerVM Virtualization Active Memory Sharing" on this topic.

This gives you the option of telling your LPARs that they have more physical memory than they actually do. Basically, you'll be paging memory to a paging device. So if all your LPARs try to do this at the same time, you could face the same kind of contention and poor performance that you'd have on a single LPAR that is memory-constrained and constantly using paging space. Therefore, you wouldn't want to do this with LPARs that all need simultaneous access to memory. However, it is helpful if you have individual LPARs that need additional memory at different times. Think of this as another way to better utilize your server. With Micro-partitioning we can better utilize CPUs; with Active Memory sharing, we can better utilize our memory across our servers. Active Memory Sharing will work across AIX 6.1, Linux and IBM i 6.1 partitions that use VIOS and shared processors and have virtual I/O. Of course, they must be POWER6 processor-based systems with the Enterprise Edition of PowerVM.

Look over all of the announcement material, including the new facts and features documentation. Then contact your business partner or IBM to help you determine how this new technology can help your business.

The Value of Being 'Well-Red'

Do you read IBM Redbooks? Some people tell me they consult Redbooks when they're seeking specific information, but that they're too busy to read them from start to finish. (And since Redbooks are typically hundreds of pages in length, they aren't generally quick reads.)

My response is that, if you look hard enough, you can find extra time for learning. For instance, do you commute on a train? That could be time spent reading. Do you watch a bunch of mindless sitcoms? Kill your television and study. Maybe there's some other time-waster you can remove from your life? Replace that with something worthwhile. Don't just punch the clock, or wait for a team lead or senior person to spoon-feed you what you should know.

To me, the Redbook, "PowerVM Virtualization on IBM System p: Managing and Monitoring," published in June, contains tons of worthwhile information.

For starters:

• Setting up a firewall on your VIO server (pages 56-58)
• Setting up NTP on your VIO server (page 73)
• Setting up additional users besides padmin on your VIO server (page 77)
• Creating vio backups (page 82)
• Backing up IVM profile data using the VIO command line with
  bkprofdata -o backup -f /home/padmin/profile.bak (page 84)
• Different ways to use the backupios command: backupios -file /mnt/backup –mksysb (page 87
• Backing up other disk structures with savevgstruct (page 93)
• Sending error logs to another server (page 127)
• Using Linux on Power tools to dynamically move memory on a running machine (page 156)
• This statement (page 170): "There are no architected maximum distances between systems for PowerVM Live Partition Mobility. The maximum distance is dictated by the network and storage configuration used by the systems, and by the fact that both systems must be managed by the same HMC. Provided both systems are on the same network, are connected to the same shared storage, and are managed by the same HMC, PowerVM Live Partition Mobility will work. Standard long-range network and storage performance considerations apply." That's interesting, even if it might not yet be practical.
• The script that creates a simple report (vios_report.ksh) of all the disks on a given VIO server (page 187)
• The tip about connecting to https:/hostname:5336/ibm/console on your AIX 6 machines (page 208). You'll get a systems director console for AIX. Take some time to try this out.

And this is just one Redbook. Take the time to read Redbooks, and any other documentation that's applicable to your job. I can promise you'll find information that you can use in your shop.

Making the Case for AIX and Power Systems

I recently received an e-mail from a mailing list that linked these documents from The Sageza Group and Forrester Research. Both reports offer information that may help non-technical personnel understand the value proposition of AIX and Power Systems.

In "The Value of PowerVM Workload Partitions New Virtualization Options in IBM AIX v6.1," Sageza focuses on on workload partitions (WPARs). I've previously covered WPARs here, here and here.

While the Sageza report is apparently accessible online, Forrester's report, "Virtualization Trends On IBM's System p: Unraveling The Benefits In IBM's PowerVM," seems to be available only through registration and subscription. From Forrester:

"IBM's PowerVM (formerly Advanced POWER Virtualization) technology has catalyzed the consolidation of server systems resources and a variety of applications workload types--both AIX- and Linux-led--as virtualized on more powerful multi-core System p servers. Evolution of IBM's virtualization stack has improved dramatically--from its early 2001 introduction of logical partitions on the first multicore POWER4-based systems--to its current PowerVM virtualization stack. In 2007, IBM's refresh to POWER6 came fast and furious: debuted with high-end System p 570 (in May), followed by the P6-based JS22 blade (November), and sweeping through the System p 520 (entry) model and System p 550 (midrange) server (January 2008). Traction for PowerVM virtualization now accounts for 70 percent of its IT customer base -- showing to what extent IBM's virtualization stack has become a shortlist contender as a systems consolidation enabler. The November 2007 release of AIX 6 added two breakthrough features--Live Partition Mobility and Live Application Mobility--further cementing IBM's advanced virtualization advantages against its Unix competitors.

"1. What's the history behind the IBM virtualization stack?
2. What is the overall business value of the System p virtualization stack?
3. What role does the POWER Hypervisor play in the PowerVM?
4. How does the POWER Hypervisor integrate with PowerVM technologies?
5. What are the benefits of micro-partitioning and the shared-processor pool?
6. What are workload partition (WPAR) and Live Application Mobility?
7. What business problems are solved with PowerVM's new Live Partition
Mobility?"

And, from Sageza:

"Many organizations have embraced virtualization to improve IT utilization and reduce the expenses associated with equipment acquisition, installation and operation. While traditional virtualization or partitioning schemes have improved IT resource utilization, reducing the number of physical servers has not reduced the number of server operating system (OS) images requiring administration and maintenance. If anything, virtualization has encouraged growth in the number of servers that support the application workloads in organizations. There is an opportunity for IT to reduce this administrative overhead to become more streamlined and cost-efficient while continuing to provide the levels of service on which organizations have become dependent.

"IBM AIX 6.1, through its support for Workload Partitions, enables organizations to rethink the way they deploy multiple workloads on a single server. While traditional approaches such as virtualization using logical partitioning provide OS isolation and independence, for many workloads, this degree of isolation exceeds the user's need and results in unnecessary administrative and operational overhead.

"WPARs offer IT managers a more cost-effective yet secured approach that meets the needs of many organizations. WPARs differ from other partitioning or virtualization schemes in that they partition server resources by the workload and share access to a single OS image. This is in contrast with the more common approach of creating a discrete operating system image to support each virtual server. By reducing the number of OS images required, the level of server software maintenance and other related IT administrative and management activity can be decreased while maintaining streamlined operational management and reduced need for physical resources.

"WPARs increase resource utilization from the typical 5-20 percent average, reduce partition creation and teardown times, and reduce the number of OS instances and associated system management workload. WPARs provide standard application environments, support mobility and templates as well as cloning, and have automated policy-based resource and workload management through the WPAR manager. Consolidating with WPARs saves floor space and reduces the power consumption and expense associated with servers and air conditioning in the data center while maintaining the one-app/one-server deployment paradigm.

"In this paper, we examine the flexibility that WPARs offer IT professionals in their virtualized UNIX server environments. In particular, we review how WPARs are different from other partitioning technologies and how WPARs complement existing environments. We discuss the capabilities and practical uses of WPARs in sample scenarios and articulate the ways in which WPARs provide an alternative to other partitioning schemes. Through AIX 6.1 and its support for WPARs and PowerVM Live Application Mobility, IT managers have greater flexibility in server configuration and can select the best approach to meet the user organization's needs while also simplifying the operational and cost efficiency of the IT environment."

PowerVM Redbook Recommendation

If you're working with PowerVM but haven't kept up with the changes, or if you're new to virtualization, then the updated Redbook, "PowerVM Virtualization on IBM System p: Introduction and Configuration" (4th edition), should be required reading. It serves as everything from an introduction to virtualization to a cookbook for setting up dual VIO servers for redundancy. Much of this post quotes directly from the Redbook, as I don't think I can say it any better than the authors have.

From the abstract:

"This IBM Redbook provides an introduction to PowerVM virtualization technologies on IBM System p servers. The Advanced POWER Virtualization features and partitioning and virtualization capabilities of IBM Systems based on the Power Architecture have been renamed to PowerVM.

"PowerVM is a combination of hardware, firmware and software that provides CPU, network and disk virtualization. The main virtualization technologies are:

* POWER5 and POWER6 hardware
* POWER Hypervisor
* Virtual I/O Server

"Though the PowerVM brand includes partitioning, software Linux emulation, management software and other offerings, this publication focuses on the virtualization technologies that are part of the PowerVM Standard and Enterprise editions.

"This publication is also designed to be an introduction guide for system administrators, providing instructions for:

* Configuration and creation of partitions and resources on the HMC
* Installation and configuration of the Virtual I/O Server
* Creation and installation of virtualized partitions

"While discussion in this publication is focused on IBM System p hardware and AIX, the basic concepts can be extended to the i5/OS and Linux operating systems as well as the IBM System i hardware.

"This edition has been updated with the new features available with the IBM POWER6 hardware and firmware."

And, from the introduction:

"The first edition of this publication was published over three years ago. Since then the number of customers using Advanced POWER Virtualization (currently named PowerVM) editions on IBM System p servers has grown rapidly. Customers use PowerVM in a variety of environments including business-critical production systems, development, and business continuity. This fourth edition includes best practices learned over the past years to build on the foundation work of the previous versions of the Redbook.

"This publication targets customers new to virtualization as well as more experienced virtualization professionals. The publication is split into four chapters, each with a different target audience in mind.

"Chapter one is a high-level introduction for those wanting a quick overview of the technology.

"Chapter two is a slightly more in-depth discussion of the technology aimed more
at the estate- or project-architect for deployments.

"Chapters three and four are aimed at professionals who are deploying the technology. Chapter three works through a simple scenario and Chapter four introduces the more advanced topics such as VLANs, Multiple Shared Processor Pools and Linux. Additionally it will introduce the techniques that can be used to provide the periods of continuous availability required in production systems."

Be sure to look for the shaded sections throughout the book, which include different sections labeled Important, Note, and Tip. Reading and understanding these will save you headaches when deploying your machines. For example, take a look at this from the SEA section:

"Note: A Shared Ethernet Adapter does not need to have IP configured to be able to perform the Ethernet bridging functionality. It is very convenient to configure IP on the Virtual I/O Server. This is because the Virtual I/O Server can then be reached by TCP/IP, for example, to perform dynamic LPAR operations or to enable remote login. This can be done either by configuring an IP address directly on the SEA device, but it can also be defined on an additional virtual Ethernet adapter in the Virtual I/O Server carrying the IP address. This leaves the SEA without the IP address, allowing for maintenance on the SEA without losing IP connectivity if SEA failover has been configured. Neither has a remarkable impact on Ethernet performance."

With this Redbook and a test environment, it wouldn't take long to better understand the topics presented.