2027 mm business class - from the first flight to the moon to cloud computing

IBM has announced the release of a new business class mainframe - zEnterprise BC12 . An amazing lineup of servers, which next year marks 50 years - half a century of development from the Apollo Moon mission to cloud computing and big data. While many processor architectures died under pressure from x86 (PA-RISC, Alpha, Itanium), mainframes were not only able to defend their right to exist, but also successfully developed ahead, winning the hearts of new customers.

S / 360, S / 390, zSeries, System z, zEnterprise ... Generations have changed, names have changed, but the key principles embodied in the first models have not changed - security, reliability, and orientation to centralized processing of a large number of tasks. Moreover, due to backward compatibility, which was one of the key principles of the development of the architecture, applications that were developed for the first models are still running even on modern representatives of this server line.

Green console native operating system
')
For 49 years, an operating system has evolved along with the mainframe, which is currently called z / OS. The interface is not very familiar. The abundance of abbreviations of tools, terminology and some other things also make you understand that the transition from the world of Linux / UNIX systems will not be too fast, but architectural features are respectful.

First, from the architectural features, I would like to immediately point out the Workload Manager. It is, of course, an essential attribute of such a serious system, which was originally planned to run various applications in a single system. Integration at all levels allows you to work wonders. For example, you can specify SLAs to execute a specific type of query in a DB2 database, and Workload Manager will force all subsystems from the application level to the computing resources (and even the storage system) to execute the query if possible at a specified time.

Secondly, built-in security tools that are also tightly integrated into the system and applications. If desired, you can even control access to individual rows of the table without additional tweaks and heavy integration with system tools.

In addition, z / OS systems are well combined into a cluster. At the hardware level, this is done thanks to the microcode, which stores in memory all the necessary cluster data (for DB2, the locking table, cache, etc.). To prevent this data from being lost in the event of server failure, it is synchronized using dedicated Infiniband channels. Thus, up to 32 systems can be combined into a cluster.

Mainframe and Linux



In addition to the native operating system on the mainframe, the well-known Linux is successfully developing. IBM-supported distros are Red Hat and SuSe, but in addition there are CentOS, Debian and Fedora.

By itself, Linux for System z is quite an ordinary Linux on a reliable piece of hardware until you reach the z / VM hypervisor. Here begin small pleasant things that can not but rejoice. For example, creating a copy of an existing system takes several minutes:
1) a text file is copied with a description of system resources, in which only the name of the new virtual server is changed in the minimal variant;
2) a new virtual server is created based on the new description (one command);
3) a snapshot of the disks is done (to simplify, a short script is written).

The whole process can be performed without stopping the cloned system and, unlike traditional x86 hypervisors, the result is predictably successful.

In general, more specifically, the System z server has two levels of virtualization. At the first level, it is divided into logical partitions (LPARs), and if desired, z / VM is installed as an additional hypervisor, which adds flexibility and allows you to run a large number of operating systems.

Of course, the z / VM hypervisor allows you to implement standard functionality, such as changing the priorities of virtual servers, adding on-the-fly new resources, and moving systems from one server to another.

Reliability in every detail

The letter Z in the name of the server line means Zero down-time - zero downtime. This emphasis on reliability is noticeable throughout. In the hardware, all components are duplicated, in addition to the standard means of memory mirroring, internal bus and external interfaces, there are even redundant processors. Moreover, IBM declares that when the main processors fail, the backup processors replace them on the fly without stopping the application.



The same level of reliability applies to all subsequent levels - reliable hypervisors, reliable operating systems, reliable applications ...

It would be strange if “zero downtime” concerned only unplanned downtime. Standard system support procedures also did not go unheeded. Upgrading the microcode and replacing hardware modules, in the case of expansion configuration or failure, pass without stopping the system. At the program level, all problems are solved by a cluster called Parallel Sysplex, which I have already talked about a little before. While one cluster node is working for others, the second can be used for any type of work. Such updates do not concern only the operating system. You can increase, for example, the version of the DBMS — DB2 v9 will still work on one cluster node, and the second one has already been raised to v10, and everything is transparent to the user ...

Productivity measured by a ruler

Traditionally, the System z platform does not participate in any public benchmarks, proudly standing apart and claiming that they have little to do with real life. Also prohibited by the measurement of gigahertz, the number of cores, and so on. The only more or less standard feature for measuring mainframe power is MIPS - millions of instructions per second.

And yet, a simple man in the street MIPSy incomprehensible and I want to measure everything with a ruler, so the key characteristics:

• Processor clock frequency - 5.5 GHz for Enterprise class, 4.2 GHz for Business class.
• Number of cores - 101 for Enterprise class, 13 for Business class.
• RAM - 3 TB for Enterprise class, 496 GB for Business class.

Personally, I do not know at the moment the server with a higher clock frequency. Moreover, the mainframe is going against the trend so far - all server-side manufacturers have gone in the direction of multi-core, reducing the clock one. One would think that this is a flaw, if it were not for one thing - most software is still licensed by core ...

Nevertheless, despite the confident standard characteristics, it is difficult to treat the mainframe as a simple fast server. There are too many non-standard things in it that are not visible to the human eye, and sometimes to the application. For example, in each mainframe there is a dedicated pool of processors that deal only with I / O operations, freeing the main processors from this unnecessary task.

With the increase in data and the growing influence of business intelligence, IBM developers began working on special optimizers for analytical queries. In the first version, it was a proprietary product of IBM Smart Analytics Optimizer - pre-installed software on a large number of x86 servers on which the analytic query was parallelized. With the purchase of Netezza, the same concept was implemented as an IBM DB2 Analytics Accelerator product. The whole process looks completely transparent to the application - DB2 independently chooses where to execute the request depending on its nature - analytical bias or transactional. As a result, OLTP and OLAP can be combined in one database without sacrificing performance.