Please enable JavaScript.
Coggle requires JavaScript to display documents.
In-memory analytics (Overview In-memory DBMS (CPU, Register, Cache, Main…
In-memory analytics
Overview In-memory DBMS
-
-
-
Main memory
- main place to store data
- lower stages only backup
- GBs, 60ns acces time, Volatile
Secondary storage
- 10 ms access time, TBs, Non-Volatile
-
Operators in IMDBs
DELETE
- lookup value ID in dictionairy
- scan attribute vectors for positions of value in ID
- Delete positions
- Re-organize positions tp fill gaps
INSERT
- without dictionairy: append new tuple to column attribute vector
- With dictionairy: add to dictionairy and get new ID
UPDATE
- update attribute vector without dictionairy
- with dictionairy: update in dictionairy in attribute vector and dictionairy, if not in dictionairy INSERT
-
In-memory Database
Resons:
- increase in analysis frequency
- analysis flexibility
- lower data latency
- more detail
- combination of OLTP and OLAP?
- become recently affordable
IMDB and new technologies
- Multicore Architecture, GPU computing (GPGPU)
- Solid State Drives (SSD)
Problems:
- access to faster storage is difficult
- different approach necessary
- capacity not enough
- no backup
- when submit transactions
RLE
- before RLE: 300,300,300,450, 500,500
- after RLE: {300,3} , {450,1}, {500,2}
-
Bit Vector
- vector for each entry which is true with 0 or 1
-