Tachyon – a hypothetical particle that moves faster than the speed of light in vacuum. In contrast to the “Tachyon”, the new project Tachyon (github) – not fiction, it is the real file system, distributed on different carriers. It is great for frameworks in the cluster, such as Spark and MapReduce.
Distributed File System (FS) guarantees the safety information in the event of a hardware failure. The failure of one carrier in the cluster does not affect the system performance and data integrity. The carrier can be replaced quickly.
Unlike other distributed file systems, Tachyon ensures not only reliable, but also high speed. For comparison, it shows 300 more bandwidth than the HDFS (Hadoop Distributed File System), due to the specific optimization and aggressive use of memory. Tachyon caches in memory work files and allows multiple processes to gain access to them, due to this and achieve high performance. This architecture is great, such as streaming media content: encoding, rendering, translation made directly from the server’s memory, never turning to the hard drive.
File system developed at Berkeley (Lab UC Berkeley AMP Lab). Tachyon system comes with software interfaces File API, which are similar in syntax to Java.
New file system can be put on a Hadoop cluster without any modifications.