Load Tests
HAKOM TSM is able to read and store large amounts of data in relational databases with high performance.
|
Benchmark |
Seconds |
Throughput rate (data point/ second) |
|
Write (1.500 ZR x 30t x 24h x 2 = 2,16 M data points) |
11 |
196.364 |
|
Read (25.000 ZR x 30t x 24h x 2 = 36 M data points) |
116 |
310.345 |
|
Parallel-Reading(5 x 5.000 ZR x 30t x 24h x 2 = 5 x 7,2 M = 36 M data points) |
110 |
327.273 |
|
Parallel-Writing (5 x 1.500 ZR x 30t x 24h x 2 = 5 x 2,16 M = 10,8 M data points) |
60 |
180.000 |
Resources:
-
1 PostgreSQL Server: 2 x Intel Xenon X5675 3.06 GHz, 8 GB RAM, Windows Server 2012
-
6 Application Server with 1 HAKOM Service: each 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
-
1 Client: 8 x Intel Xeon X5675 3.06 GHz, 16 GB RAM, Windows 10
Scaling-Tests
The following benchmarks show horizontal and vertical scaling effects with HAKOM TSM Services.
Data was written and read for 1,000 time series in 15-minute interval (96) over 31 days (2,976,000 values) using single and bulk requests. In the case of bulk requests, one bulk request was sent in parallel for every 100 time series, so that these tests also simulate simultaneous access by 10 users with 100 time series each.
HAKOM Standard License vs. HAKOM Performance License (incl. service parallelization)
Standard vs. Performance column shows the percentage improvement of Performance License compared to Standard (1-Performance/Standard%).
|
Benchmark for (n=10) |
App Server |
Standard |
Performance |
Standard vs. Performance |
|---|---|---|---|---|
|
Write Singe-Request |
1 |
00:27,649 |
00:10,954 |
60,38% |
|
Read Singe-Request |
1 |
00:08,625 |
00:07,078 |
17,94% |
|
Write Bulk-Request |
1 |
00:35,103 |
00:11,885 |
66,14% |
|
Read Bulk-Request |
1 |
00:07,127 |
00:05,344 |
25,02% |
|
Write Bulk-Request |
2 |
00:26,335 |
00:09,811 |
62,75% |
|
Read Bulk-Request |
2 |
00:05,620 |
00:03,695 |
34,25% |
Resources:
-
1 PostgreSQL Server: 2 x Intel Xenon X5675 3.06 GHz, 8 GB RAM, Windows Server 2012
-
1 to 2 Application Server with 1 HAKOM Service: each 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
-
1 Client: 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
Horizontal Scaling Effects with 1 to 4 App Servers
The following results were obtained with Performance License over 1 to 4 App Servers.
The Δ X vs. Y Server columns show the percentage improvement of X number of servers vs. Y number of servers.
|
Benchmark for (n=10) |
1 App Server |
2 App Server |
4 App Server |
Δ 1 vs. 2 Server |
Δ 2 vs. 4 Server |
Δ 1 vs. 4 Server |
|---|---|---|---|---|---|---|
|
Write Bulk-Request |
00:11,885 |
00:09,940 |
00:09,232 |
16,37% |
7,12% |
22,32% |
|
Read Bulk-Request |
00:05,344 |
00:03,757 |
00:02,907 |
29,70% |
22,62% |
45,60% |
Resources:
-
1 PostgreSQL Server: 2 x Intel Xenon X5675 3.06 GHz, 8 GB RAM, Windows Server 2012
-
1 to 4 Application Server with 1 HAKOM Service: each 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
-
1 Client: 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
Horizontal and Vertical Scaling in Comparison
The following table compares scaling effects between the following constellations when reading 1,000 time series in 15 minute interval over 31 days:
-
2 Engines with 8 Cores, 16 GB RAM and 2 HAKOM Services (2 x 8 Cores)
-
1 Engine with 8 Cores, 16 GB RAM and 1 HAKOM Service (1 x 8 Cores)
-
2 Engines with each 4 Cores, 8 GB RAM and each 1 HAKOM Service (2 x 4 Cores)
-
1 Engine with 4 Cores, 8 GB RAM and 1 HAOM Service (1 x 4 Cores)
|
(x / y) |
x → |
2 x 8 Cores |
2 x 4 Cores |
1 x 8 Cores |
1 x 4 Cores |
|---|---|---|---|---|---|
|
y ↓ |
|
00:02,248 |
00:02,286 |
00:02,692 |
00:03,125 |
|
2 x 8 Cores |
00:02,248 |
- |
101,69% |
119,75% |
139,01% |
|
2 x 4 Cores |
00:02,286 |
98,34% |
- |
117,76% |
136,70% |
|
1 x 8 Cores |
00:02,692 |
83,51% |
84,92% |
- |
116,08% |
|
1 x 4 Cores |
00:03,125 |
71,94% |
73,15% |
86,14% |
|
Resources:
-
1 PostgreSQL Server: 2 x Intel Xenon X5675 3.06 GHz, 8 GB RAM, Windows Server 2012
-
1 to 4 Application Server with 1 HAKOM Service: each 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
-
1 Application Server with 1 up to 2 HAKOM Services: 8 x Intel Xeon X5675 3.06 GHz, 16 GB RAM, Windows 10
-
1 Client: 2 x Intel Xeon X5675 3.06 GHz, 8 GB RAM, Windows 10
Compression Break-Even-Point
In HAKOM TSM time series values can be stored compressed. Thereby several values are stored in blocks with only one block timestamp, resulting in an significant reduction in the amount of data and database performance. Read more: Time Series Compression
The following evaluation shows from which % of the value changes of a compression block the performance becomes better with compression than without compression.
|
|
Compressed |
25% of the block |
50% of the block |
75% of the block |
100% of the block |
|---|---|---|---|---|---|
|
Write Bulk-Request |
yes |
00:01,469 |
00:01,602 |
00:01,519 |
00:01,454 |
|
Write Bulk-Request |
no |
00:01,248 |
00:01,421 |
00:01,514 |
00:01,632 |
|
Read Bulk-Request |
yes |
00:00,870 |
00:00,932 |
00:00,932 |
00:00,947 |
|
Read Bulk-Request |
no |
00:00,870 |
00:00,932 |
00:00,932 |
00:00,995 |