Hardware Sizing

Hardware Sizing (Standalone Install)

Small Tier - 16 Core, 128G RAM (r5.4xlarge / E16s v3)

Component	RAM	Cores
Web	2g	2
Postgres	2g	2
Spark	100g	10
Overhead	10g	2

Medium Tier - 32 Core, 256G RAM (r5.8xlarge / E32s v3)

Component	RAM	Cores
Web	2g	2
Postgres	2g	2
Spark	250g	26
Overhead	10g	2

Large Tier - 64 Core, 512G RAM (r5.16xlarge / E64s v3)

Component	RAM	Cores
Web	4g	3
Postgres	4g	3
Spark	486g	54
Overhead	18g	4

Important  Collibra DQ requires a limit of 2TBs for large tier jobs. For DQ jobs that exceed 2TBs, you must filter down columns or rows.

Estimates

Sizing should allow headroom and based on peak concurrency and peak volume requirements. If concurrency is not a requirement, you just need to size for peak volume (largest tables). Best practice to efficiently scan is to scope the job by selecting critical columns. See Performance Tuning for more information.

Bytes per Cell	Rows	Columns	Gigabytes	Gigabytes for Spark (3x)
16	1,000,000.00	25	0.4	1.2
16	10,000,000.00	25	4	12
16	100,000,000.00	25	40	120
16	1,000,000.00	50	0.8	2.4
16	10,000,000.00	50	8	24
16	100,000,000.00	50	80	240
16	1,000,000.00	100	1.6	4.8
16	10,000,000.00	100	16	48
16	1,000,000,000.00	100	1600	4800
16	100,000,000.00	100	160	480
16	1,000,000.00	200	3.2	9.6
16	10,000,000.00	200	32	96
16	100,000,000.00	200	320	960
16	1,000,000,000.00	200	3200	9600

Cluster

If your program requires more horsepower or (Spark) workers than the example tiers above which is fairly common in Fortune 500 companies than you should consider the horizontal and ephemeral scale of a cluster. Common examples include Amazon EMR and Cloudera CDP. Collibra DQ is built to scale up horizontally and can scale to hundreds of nodes.