Scratch (computer)
The scratch memory (often Scratch -Disk from English. Scratch , scratch) is a nonvolatile separate memory area only for temporary files is used. While computers used to have separate hardware for scratch memories, the term is now only used in niche areas of big data , such as high-performance computing , video processing or image processing .
Historical
Scratch Space was useful for solving large tasks, such as sorting a long list that did not fit in the main memory of a computer, on writable storage media . While when using punched cards, the physical storage medium itself with its position in the computer serves as a swap memory (for example in the punched card sorter ), individual scratch tapes were already shown on computers with tapes , which only served as temporary storage, as later hard drives did . The rapid wear and tear of this buffer prompted the name scratch .
Scratch space in personal computing
Scratch space is used by programs that require more memory than the computer has available in memory . Unlike when the operating system uses a swap file , the deliberate use of such a file from a userspace program prevents swapping , which slows down the entire computer. The main difference is that the programmers of the user program had the large memory requirement in mind and programmed the use of a non-volatile swap memory.
As a rule, a separate partition on the hard drive is used for scratch space . On mobile devices , a single folder on a mass storage medium, such as an SD card , is often chosen. An example of a program that uses Scratch Space is Adobe Photoshop .
Since the use of scratch memories in personal computing is not managed by the operating system, there is no standard way of handling full scratch memories.
Scratch space in high-performance computing
Having a very large scratch disk is a primary characteristic of a high performance computer . This is typically a high-speed , distributed network storage that is spread across thousands of hard drives. As a rule, the scratch disk is distinguished from other non-volatile storage media in that the content is not backed up or deleted by system administrators or automatically, sometimes without warning.
The scratch disk is usually the main hard drive in high-performance computers and represents the majority of the available memory. The SuperMUC at the Leibniz data center has around 15 petabytes of network memory without backup, some of which is used as scratch memory (deletion without Announcement) and a part as work memory (no deletion). With a typical hard drive size of 1 terabyte, the scratch memory is distributed over approx. 15,000 hard drives. In contrast, there are only 3.5 petabytes of network storage that have a local replication, so that 7 petabytes of storage have to be kept available. As a file system for scratch comes GPFS from IBM for use. Another example is the high-performance computer Hazel Hen at the high-performance computing center in Stuttgart , which uses the Luster file system to manage its 10 petabyte scratch memory and an allocation system that is similar to the allocation of CPU resources and is provided with an expiration time to prevent that the scratch file system fills up to the brim. In order to use the memory, users must therefore apply for a quota in which they reserve a certain amount of memory for a period of time.
Individual evidence
- ↑ Definition of: scratch tape. In: Encyclopedia. PC magazines. From PCmag.com, accessed September 30, 2019.
- ↑ SuperMUC Petascale System. In: System Description. Leibnitz data center. At LRZ.de (English), accessed on September 30, 2019.
- ↑ Cray XC40 (Hazel Hen). High-performance computing center Stuttgart. On HLRS.de (English), accessed on September 30, 2019.
- ↑ Workspace mechanism. HLRS Wiki. At kb.HLRS.de (English), accessed on September 30, 2019.