With the advent of cloud, in general there is less talk about mainframe computers than some time ago. Yet i mainframe they still exist today and indeed are actively used “behind the scenes” for critical applications (one might say mission critical) in multiple contexts: it is estimated that 45 of the top 50 banks in the world, 8 of the top 10 insurance companies, 8 of the top 10 telecommunications operators make use of mainframes. The concept of mainframe is often, wrongly, connected to outdated systems, now consigned to the history of computer science. In reality, as we mentioned, they are alive and well outperforming most modern servers.
What is a mainframe and how does it work?
A mainframe is a type of computer High performance designed to handle large volumes of data and provide reliable, scalable and secure services for businesses, financial institutions, governments and other organizations. Mainframes are one of the earliest forms of computers and are still widely used in critical industries where reliability, security and performance are paramount.
The term mainframe described the former big computers centralized networks that dominated the computing landscape from the 1950s onwards. When the era of personal computing was yet to begin, they wanted to differentiate those large systems size from devices intended for personal use that would later hit the market.
At the dawn of the information age, IBM had several competitors all proposing their respective mainframe solutions: Big Blue, however, dominated the industry with 80% market share. Today, in fact, mainframe is synonymous with IBM and after 50 years its systems continue to be current. Let’s see why. Meanwhile, you can access an interactive demo created by IBM.
The characteristics of a modern mainframe have obviously changed dramatically from what was on offer in the 1950s and 1960s. Today, mainframes are still powerful and reliable, but they have been updated and optimized to meet the needs of a wide variety of business applications and institutional.
A modern day mainframe can accommodate up to 240 CPU of level server40 TB of RAM with ECC functionality (Error Correction Code) and many petabytes of redundant flash-based storage. These are therefore systems designed to process large amounts of critical data while maintaining a uptime of 99.999%. This equates to approximately five minutes of downtime per year of operations. Mainframes are designed to provide both redundancy That throughput high.
Redundant Computing: The flagship of mainframes
We talk about redundancy to refer to an architecture or system that uses components in a configuration that ensures greater reliability and availability of services. The goal of redundant processing is to prevent a broken down on a single component can negatively impact the operation of the rest of the system. In this way, with a parallelized approachthe system can remain operational and functional even in the presence of failures.
Mainframes are designed in partitions independent, each with different RAM, storage, CPU and even operating systems. Applications can continue to run normally even when some of the partitions are subjected to operations maintenance: operating system patches, application of updates of various types and interventions on the hardware.
In this regard CPU, RAM and data storage units are hot swappable: If a component fails, it can be pulled out and replaced without requiring a mainframe shutdown. That’s why companies can do this business-as-usual without having to support any service interruption provided to end users.
Weapon: IBM has its own CPU through the mainframe system
In August 2021, IBM presented what remains the most updated version of the CPU per mainframe. Call A weaponthe IBM CPU was made for the mainframes of z16 series using Samsung’s 7nm manufacturing process and is developed on a die which measures a good 530 mm2. It is a CPU with 8 physical cores that become 16 logical cores thanks to the integrated SMT2 support.
Telum can go to clock speed above 5 GHz and uses 256 MB of cache divided into eight groups of L2 caches of 32 MB each. It is also the first IBM processor to supportacceleration in hardware for the inference operations related to the processing of applications for theartificial intelligence. The AI functions they support both inference processing and the machine learning.
The architecture of the IBM Telum processor
Unlike processors from Intel and AMD as well as ARM SoCs, IBM mainframe processors are based on the z/Architecture. This architecture was introduced by IBM in 2000 and is the foundation upon which all mainframes are built IBM Z subsequent. It is optimized to perform high-speed transactional operations and handle large volumes of data efficiently. Due to its robustness and advanced features, z/Architecture is widely used in large organizations and industries where maximum reliability e safety of corporate operations.
Cache L2 and L3
Speaking of cache we said that each Telum core consists of 32 MB of cache L2. Removing a row from the L2 cache moves it to another L2 cache and marks it as belonging to the L3 cache. Eight L2 caches can then be combined into one virtual L3 cache shared 256MB accessed by any of the 8 cores.
To make a comparison, take a chiplet as an example AMD Zen 3: In this case, each of the 8 cores has 512KB of L2 cache and a total of 32MB of L3 cache.
By removing a line from the L3 cache, it can find a place in another core’s cache, thus creating virtual L4 cache. Combining the caches provides something like 8192 MB of virtual L4 cache. IBM already claimed at the time that Telum improves overall performance per socket by more than 40% over the previous z15 CPU design.
Plus Telum to form the mainframe system
In the context of mainframes, a drawer (drawer) is a physical hardware module that contains specific components, such as processors, memory, input/output boards, and other devices that are essential for system operation. We said that Telum consists of 8 physical cores per single chip. Two Telum chips are used for each socket; four sockets for each drawer. In turn, 4 drawers make up a system. As mentioned in the introduction, a mainframe can support up to 250 cores, 190 of which are user controlled. The I/O is handled by two controllers PCIe 4.0.
Logical partitions and storage on the mainframe
We said that the mainframe is organized into partitions. each logical partition (LPAR) can run a separate, isolated instance of Z/OS, the mainframe operating system developed and maintained by IBM. However, the various LPARs can also run other operating systems, such as Linuxand have completely separate hardware resources.
LPARs are also separated through the use of a robust system permissions: A group of users can have access to the LPAR used for testing but not to the LPARs used in production.
The protocol parallel sysplex from IBM allows you to couple different mainframes or multiple LPARs on the same system to manage i workloadto activate communications in the various directions and for recovery needs or failover.
As for it data storage, storage arrays are not attached to the mainframe. The latter can be structured by matching hard disk e SSD (possibly NVMe) and thus forming an overall capacity that can widely exceed, today, 6 Petabytes and reach 48 GB/s of troughput. For comparison, just remember that a single performance NVMe drive maxes out at around 7GB/s.
The IBM Z/OS operating system
Released…