Main Specifications of CarbonData¶
Main Specifications of CarbonData¶
Entity | Tested Value | Test Environment |
---|---|---|
Number of tables | 10000 | 3 nodes. 4 vCPUs and 20 GB memory for each executor. Driver memory: 5 GB, 3 executors. Total columns: 107 String: 75 Int: 13 BigInt: 7 Timestamp: 6 Double: 6 |
Number of table columns | 2000 | 3 nodes. 4 vCPUs and 20 GB memory for each executor. Driver memory: 5 GB, 3 executors. |
Maximum size of a raw CSV file | 200GB | 17 cluster nodes. 150 GB memory and 25 vCPUs for each executor. Driver memory: 10 GB, 17 executors. |
Number of CSV files in each folder | 100 folders. Each folder has 10 files. The size of each file is 50 MB. | 3 nodes. 4 vCPUs and 20 GB memory for each executor. Driver memory: 5 GB, 3 executors. |
Number of load folders | 10000 | 3 nodes. 4 vCPUs and 20 GB memory for each executor. Driver memory: 5 GB, 3 executors. |
The memory required for data loading depends on the following factors:
Number of columns
Column values
Concurrency (configured using carbon.number.of.cores.while.loading)
Sort size in memory (configured using carbon.sort.size)
Intermediate cache (configured using carbon.graph.rowset.size)
Data loading of an 8 GB CSV file that contains 10 million records and 300 columns with each row size being about 0.8 KB requires about 10 GB executor memory. That is, set carbon.sort.size to 100000 and retain the default values for other parameters.
Table Specifications¶
Entity | Tested Value |
---|---|
Number of secondary index tables | 10 |
Number of composite columns in a secondary index table | 5 |
Length of column name in a secondary index table (unit: character) | 120 |
Length of a secondary index table name (unit: character) | 120 |
Cumulative length of all secondary index table names + column names in an index table* (unit: character) | 3800** |
Note
* Characters of column names in an index table refers to the upper limit allowed by Hive or the upper limit of available resources.
** Secondary index tables are registered using Hive and stored in HiveSERDEPROPERTIES in JSON format. The value of SERDEPROPERTIES supported by Hive can contain a maximum of 4,000 characters and cannot be changed.