A quick overview of IBM Power Systems

A year ago, as a member of the academic. IBM programs, I took an IBM AIX 6 Jumpstart for UNIX Professionals course at IBM's Moscow office. The instructor covered most of the topics for an experienced Linux or UNIX administrator to sit behind AIX and start working, and also told about the main points of the IBM Power Systems server line. As it turned out, these are quite interesting Enterprise-level systems that provide complete business systems, including built-in virtualization, monitoring, diagnostics, etc., which are also supported by the operating system.
I do not claim to be an exhaustive description of the architecture, the article is rather an excursion into technology for general development.

And so, as I said, System P is a line of IBM servers built on the basis of Power processors (RISC architecture). The main purpose is an application server for business systems: a database server, corporate correspondence (for example, Lotus) and various middleware (middleware) such as WebSphere.
There are 3 operating systems on the Power servers: native AIX, IBM i and, of course, Linux.
AIX is one of the oldest UNIXes, known for its reliability, stability, flexibility and ease of management (mainly achieved through SMIT - a console system configurator, with which you can do a lot of things without bothering with command line options). The fact that RPM packages from Linux are being put to me was a shock to me when studying this system! Under its architecture, of course, but still. It was not difficult to put on AIX native mc: I downloaded the RPM package for the ppc architecture, downloaded the dependencies, and the RPM team RPM I put it all.
IBM i is a separate operating system that has little resemblance to all known families. I don’t know much about it, but friends from the bank who service AS / 400 with IBM i, speak of it as one of the most productive operating systems for DB2 DBMS.
Linux for Power is officially supported only from two vendors: RedHat, with their RHEL, and Novell, with their SLES. How many did not look for a sign describing the functions of System P, which are not supported by Linux, did not find, but I know that there are such. For example, the fact that ext3, unlike JFS2 from AIX, does not support reducing the size of a partition without unmounting, is very unpleasant, so operations with a reduced LVM partition lead to the downtime of any services.

The main technological “chips” of System P, distinguishing this line from the whole variety: developed hypervisor, “resources on demand”, flexibility and mobility in building and changing infrastructure, energy efficiency, clustering support to ensure high availability. Prices, of course, at first glance bite, but in the end all of the above characteristics and technologies fully pay back the initial costs. Now in order:

Virtualization tools
This is probably the most interesting part: in Power systems, all operating systems work in virtual environments based on the hypervisor. Those. The server can contain several isolated from each other "guests" working simultaneously and sharing resources. Each logical (virtual server) is called an LPAR (Logical Partition Access Resources), into which the operating system is installed. The maximum number of LPARs on a server depends on the number of processors and memory: you can create a maximum of 10 LPARs per processor, allocating a minimum share of 0.1 processor resources and at least 256 MB of RAM to each LPAR. There is also a theoretical limitation: for 795 models (the last in the line), you can create a maximum of 254 LPAR. Thus, it is possible to optimize the server load by dividing the processor pool into virtual servers, i.e. to some LPAR give 0.5 processor, and more loaded - 1.2.
Also, each LPAR is allocated disk space and network cards. This is a special server called VIOS (Virtual I / O Server). It consolidates resources: network cards, disk space SAN or local drives, and, according to the user's settings, gives it all to virtual machines.
But the most pleasant thing is the DLPAR (Dynamic Logical Partitioning) technology: it allows you to dynamically change the amount of resources (processor shares, memory, I / O interfaces) allocated for the LPAR without stopping the system. Those. if I looked at loading partitions and saw that one uses 90% of resources on average, and the other 30%, I can safely bite off resources from the second partition without stopping the system and give these resources to the first.
All management of a hypervisor falls on one of two systems: IVM (Integrated Virtualization Manager) or HMC (Hardware Management Console). IVM comes in a package to the VIOS, and provides a convenient web interface for managing the entire server. The HMC is already a separate machine with a stripped-down version of linux, which will be required to manage multiple servers. It also has a convenient web interface and all servers can be managed right here. The web interface, of course, can be exchanged for the console, the system provides both options.
')
Resources on request
It is rather not a feature of the piece of iron itself, but a feature of licensing. It allows you to buy a server with a large number of resources, leaving some inactive, and paying only a fraction of their cost for them, and then, when the server is already in short supply, you can pay IBM for activation and add the necessary resources. You can not activate all the resources on an ongoing basis: you can activate them for a while and pay by the minute for the resources used. This allows you to add resources at peak times, and not pay for them when the server is relatively idle.
Another rather pleasant moment: if, for example, the working processor fails, and there is not an activated processor, then the worker turns off, and the inactive turns on, without additional payments.

Mobility
When building the infrastructure, you can use the PowerVM Enterprise Edition. Its main function: Live Partition Mobility. This technology allows servers to send each other LPARs without disconnecting them, even under load. Of course, it is advisable to use fast channels, SAN arrays, etc., so that all this is fast and customers do not even notice. At IBM, we were shown how this happens: a loaded LPAR “moved” from one building to another in seconds, and began working there.
Once with the students they conducted an experiment on a virtual box: they took a virtual machine and tried to transfer the virtual machine to another computer by LAN. Then this function was new, only appeared in the virtual box, and the most interesting thing is that it worked. However, as it turned out, only if the virtual machine is not loaded. As soon as I loaded it, the hypervisor performed 10 iterations and became buggy. Of course, 100 megabit local storage and general storage in the form of an SMB resource in the class does not compare with the SAN array and IBM 10-gigabit optics, but it was worth a try.

Energy efficiency
Power servers have built-in power and cooling management systems that help save energy and reduce costs, provide better performance in terms of watts, and Advanced POWER Virtualization allows you to consolidate non-fully-loaded servers that waste energy. In addition, EnergyScale technology can provide unique power management capabilities. These features allow users to measure the power consumed by system components and configure policies so that the server environment is energy efficient.

High availability
Power-based servers can be clustered, providing high availability and cluster multi-processing. This technology was called PowerHA. PowerHA provides automatic fault detection, diagnostics, application recovery, and node reintegration. Together with the appropriate software support, PowerHA can provide simultaneous data access for parallel processing applications, thus offering excellent scalability.

Conclusion: POWER-based servers have very interesting technological innovations and allow you to build a very flexible and sustainable infrastructure.

From the literature on Power I advise you to study redbooks Finding a system to see how it works is quite difficult: the systems are expensive and rather rare, at least for us. Purely human curiosity can be satisfied by watching the video on youtube, or in the documentation on the IBM developerworks website. There is also a great Russian site dedicated to Power Systems and AIX: aixportal.ru .