IBM / Mainframe
Mainframes
History
- System between Hardware and Application
- first use of an Architecture on the IBM System 360
Usage
Areas
- Banking (UBS, …)
- Airlines
- Super Markets (Rewe)
- Insurances
- …
Transactions
IBM mainframes runs ~70% of all world transactions by value
- 77 of top 100 banks in world use IBM mainframe
- revenue & installed capacity increasing
Technology Trends
Observations:
- Core Frequency stagnated -> #cores started increasing
- #Transistors growing
- Single Thread Performance (slowly) increasing
- Demand and energy consumption has drastically increased
- AI made this much worse.
- Dennard scaling has stopped recently, which lead to increased consumption
⇒ We cannot continue to build computers like before.
How is IBM different?
Fails & How to address
- Hard Fail
- Bugs Pre- / post-Silicon Verification & Formal Verification
- Manufacturing Issue Testing, Redundancy & Margins
- Aging Margins
- Soft Fail
- Random Event redundancy & better technology and design
- Electrical Noise electircally robust design point
Mainframe VS Small Machines
- costs are reduced with large system (Example: Prime Video)
- many small systems need large & complex software
Specifications
Depending on configuration: up to
200 cores
Single-Thread 5.2 GHz
40 TB RAM
2 GB of L4 Cache
Example Workload: 25 billion encrypted transactions/day (Redundancy excluded!)
Easy to Use
- Many technologies are integrated in Hardware and firmware stack
- developer doesn’t need to care about that.
Efficient
- Chips are good in scaling
- IBM Z can efficiently distribute workload onto multiple cores.
- effective for peak load
- Support for Garbage Collection, AI, Encryption, and more
Redundancy
- redirection of resources
- still works with partial hardware missing
- with HW functionality, archives complete software runtime without interruption
- The System can run uninterrupted by normal affairs.
- Everything exists twice
- Unmatched reliability by using redundant elements and error correction, e.g. in L2-caches
- Mainframe don’t have a single point of failure
- Replace Hardware while still running
Secure and Reliable
- data encrypted automatically
- Regulators are eager for secure and reliable systems.
- You need to be able to prove to regulators that you provide these requirements.
- e.g: prevent example Buffer Overflows (by blocking it on a lower level).
- Post-Quantum Secure
full backwards compatibility
- Preservation of Investments
- Requirements in banking stayed basically the same
- Programs should work for decade