2 ms·
Bigtable is exactly this. An entire database is saved in a small number of files which are are append-only (and therefore can be written fully linearly). Read
by londons_explore 7y ago
Bigtable is exactly this. An entire database is saved in a small number of files which are are append-only (and therefore can be written fully linearly).
Reads of a single record usually only read a single disk block. In-memory indexes to locate the block are fairly small, even for 1B+ records.
Reads of a range of records do N range reads of blocks, where N is typically < 5, and the number of bytes read compared to the number of bytes returned to the user is typically <10% overhead.