Shulou(Shulou.com)06/03 Report--

First, the testing background

Due to the well-known world economic and trade situation, having a comprehensive and self-controlled IT solution has become an important task in front of the country. However, at present, the performance of domestic chips is still relatively weak, and most of them can only be used in marginal businesses such as OA, so it is necessary to increase the cost to expand the cluster scale when it is used in key businesses, which will further push up the application threshold of the national production technology scheme, which is not low in cost due to the lack of ecological maturity, and restrain the scope of popularization and application of national production technology.

SPL computing database is a new technology with complete independent property rights from theoretical model to code implementation. it adopts innovative algebraic system, which allows programmers to implement higher-performance and low-complexity algorithms, so as to achieve better computing performance than traditional relational databases. Therefore, the hardware can be compensated by software, so that the domestic chips with lower performance can adapt to the application scenarios where foreign chips must be used without expanding the size of the cluster, and the performance is improved to a certain extent.

Second, test methods

Using the internationally recognized database performance testing tool TPCH, it is compared with Oracle12.1.0, an outstanding representative of foreign database products, and another domestic database (hereinafter referred to as domestic database).

TPCH is a set of test benchmark for database decision support capability published by TPC (Transaction Processing Performance Council) transaction performance Committee. It examines the comprehensive processing capability of the database by simulating complex business-related queries in the database and obtains the response time of database operations.

A database model is defined in the TPCH benchmark model, and the capacity can be selected from among the eight levels of 1GB~10000GB. The database model includes 8 data tables of CUSTOMER, LINEITEM, NATION, ORDERS, PART, PARTSUPP, REGION and SUPPLIER, involving 22 complex SQL query statements.

TPCH testing has a wide range of authority and recognition, involves rich content, can avoid one-sidedness, and ensure the persuasive conclusion of the test.

Traditional databases are queried in SPL language. In SQL database, the SQL provided by TPCH needs to be translated into SPL script to run.

Third, the test environment

This test is run on five independent servers with different configurations, named intel2670, intel3014, Godson, Feiteng FT1500 and Feiteng FT2000. The hardware configuration is shown in the following table:

Name CPU memory hard disk type Intel26702 intel2670 CPU, main frequency 2.6 G, total 16 core 128GSSD solid state disk Intel30142 intel3014 CPU, main frequency 1.7 G, total 12 core 64 G Godson 2 CPU, total 8 core 64GFT1500 Feiteng CPU, total 16 core 32GFT2000 Feiteng Feiteng CPU, total 64 core 256G

The domestic database is only tested on Godson, Oracle is only tested on intel, and SPL is tested on all five servers.

IV. Data preparation

The test data are generated by the data generation program provided by TPCH, which produces 100G and 200G test data. 100G data is used on four servers of intel3014, Godson, Feiteng FT1500 and Feiteng FT2000, and 200G data is used on two servers of intel2670 and Feiteng FT2000.

After the original text file data is generated by TPCH, it needs to be imported into the database table for database software query. The data used by the SPL database is stored in the hard disk group table file and also needs to be generated by the original text data conversion. The following table shows the time it takes to convert to the required test data (there are eight tables, comparing only three tables with a larger amount of data).

The data table name is LINEITEMORDERSPARTSUPP100GOracle (intel3014) 09:40:5901:31:3400:59:02SPL Group Table (intel3014) 01Part 4000Groupe 17VOO 4800VOV 47 homemade database (Godson) 01:55:3700:17:4100:10:18200GOracle (intel2670) 37:47:2807:43:5403:12:30SPL Group Table (intel2670) 012VOV 5700 VLV 25GOOG 200017REV 23

As you can see, the import of Oracle data takes a very long time. The import time of SPL is the same as that of domestic database, which has several times better performance than Oracle.

5. Test results

The three kinds of software tested support parallel query, that is, multi-thread is used to coordinate and complete a query task. However, the number of parallel paths is not as many as possible, so it is necessary to measure the number of parallelism with the fastest query speed on each test server. We tested with the first query task of TPCH, and the parallel query results on each test service are as follows:

TPCH1 Test response schedule (seconds)

Data volume server parallel number 12481216100GIntel3014Oracle570356219170131141SPL group table 33617491463840 Godson domestic database 1620961553507579694SPL group table 1521

525275284

FT1500 Feiteng SPL Group Table 71136719895

62 number of server parallelism 81632404864FT2000 Feiteng SPL group table

3722192140200GIntel2670Oracle318316318

SPL Group Table 634041

FT2000 Feiteng SPL group table

thirty-six

The parallel number corresponding to the red number in the above table is the best parallel number on each test server, so this parallel number is used in the subsequent query tests of TPCH.

The test results are as follows:

Total 22 query test response schedules for TPCH (seconds)

TPCH

Serial number

100g data 200G data

Intel3014

(12 parallel)

FT1500

(16 parallel)

FT2000

(40 parallel)

Godson

(8 parallel)

Intel2670

(16 parallel)

FT2000

(40 parallel)

SPLOracleSPLSPLSPL domestic database SPLOracleSPL1381316219275507403253624278618247873133212223322974451235823541520727188917902145443534225362472176125463456121352266075711352127251843220917002249630837192464893161129485809682341256551710668563613510212153522991634344934211533126291659631112201847238173647524645513573711485335220913510313514221576512142500653683815181556026103506613584616101319125310514712217171654891009634034919181634435131632382259662619151546512137518603452920151755711110594554422021273262221909013349191790398221948372799139309949 total 51635631232691375626601103587771319

VI. Conclusion

Make the test results into a statistical diagram:

Some conclusions can be seen as follows:

1. Using the same TPCH test data on the same test machine, comparing the total time of 22 queries, the SPL using the new algorithm is 5-7 times faster than Oracle, and the innovative theory and algorithm are indeed effective.

two。 On Feiteng FT1500, it is nearly 3 times faster than Oracle on intel3014, and more than 5 times faster than Oracle on intel2670 on Feiteng FT2000. The SPL database on Feiteng chip far exceeds the computing performance of Oracle on Intel, and there is no problem when it is used in key business.

3. Godson is relatively weak, but with the support of SPL, it can basically match the speed of oracle on intel3014 (close to 95%), and can basically be used in critical business.

4. The performance of the domestic database on Godson is poor, which can only reach about 1/7 of the Oracle on Intel3014, and basically does not have the ability to apply to critical business.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.