Hologres:Test results

Test environment

Table row counts

Instance specifications

OLAP query performance

All queries run on Hologres internal tables. Execution time is measured in seconds — lower is better. Doubling the compute resources (64 CUs to 128 CUs) reduces execution time proportionally.

V3.1.3 total: 64 CUs — 10.4 s | 128 CUs — 7.05 s. Compared to V2.2.19, V3.1.3 reduces total query time by 12% at 64 CUs and 8% at 128 CUs.

Key/Value point query performance

All queries run on Hologres internal tables. Software version: r1.1.42. Each scenario runs continuously for 5 minutes. Queries per second (QPS) scales linearly as instance size increases.

Scenario 1 — single-key lookup

SELECT O_ORDERKEY, O_CUSTKEY, O_ORDERSTATUS, O_TOTALPRICE,
       O_ORDERDATE, O_ORDERPRIORITY, O_CLERK, O_SHIPPRIORITY, O_COMMENT
FROM   public.orders_row
WHERE  o_orderkey = {parameter};

Scenario 2 — 9-key IN-clause lookup

SELECT O_ORDERKEY, O_CUSTKEY, O_ORDERSTATUS, O_TOTALPRICE,
       O_ORDERDATE, O_ORDERPRIORITY, O_CLERK, O_SHIPPRIORITY, O_COMMENT
FROM   public.orders_row
WHERE  o_orderkey IN (
    {parameter1}, {parameter2}, {parameter3},
    {parameter4}, {parameter5}, {parameter6},
    {parameter7}, {parameter8}, {parameter9}
);

Data write performance

All results use 64 CUs and measure time in seconds to import data into Hologres internal tables.

The COPY statement imports one file into one table at a time. Concurrent import is unavailable with COPY.

Key findings:

• Virtual private cloud (VPC) import is significantly faster than Internet import because COPY performance is bandwidth-bound.

• Importing via a MaxCompute foreign table is faster than COPY over both network paths.

The bar chart below compares import times across all three methods. Blue bars represent Internet COPY, green bars represent VPC COPY, and gray bars represent MaxCompute foreign table. Lower values indicate faster import.

The x-axis represents the table name. The y-axis represents import time in seconds.

Data update performance

All results are based on upsert operations on Hologres internal tables. Software version: r1.1.42. Each scenario runs continuously for 5 minutes. QPS scales linearly as instance size increases.

INSERT INTO public.orders_row(
    o_orderkey, o_custkey, o_orderstatus, o_totalprice,
    o_orderdate, o_orderpriority, o_clerk, o_shippriority, o_comment
) VALUES (
    {parameter}, 1, 'demo', 1.1, '2021-01-01', 'demo', 'demo', 1, 'demo'
)
ON CONFLICT(o_orderkey)
DO UPDATE SET (
    o_orderkey, o_custkey, o_orderstatus, o_totalprice,
    o_orderdate, o_orderpriority, o_clerk, o_shippriority, o_comment
) = ROW(excluded.*);

Test environment

Table row counts

Instance specifications

OLAP query performance

All queries run on Hologres internal tables. Execution time is measured in seconds — lower is better.

V4.0.4 total: 69.99 s — 8% faster than V3.1.3 (76.27 s) and 31% faster than V3.0.5 (101.16 s).

Test environment

Hologres instance

Flink configuration

Test results

During the test window from 19:35 to 19:50 on April 8, 2022, the peak records per second (RPS) reached 580,234 and the minimum RPS was 357,729.