1 Verification Environment

1.1 Hardware Configuration for the Verification

1.2 Software Configuration for the Verification

1.3 Networking Planning for the Verification

Fig. 2: GBase 8a MPP Networking Planning Architecture

2 Verification Method

Step 1: Configure 6 SAS UM511a SSDs Raid5.

Step 2: Create Database and Shardings

 Use TPC-H tool to evaluate the database analysis and processing capabilities, create the database and 8 TPC-H built-in shardings in advance;

 Use TPC-DS tool to evaluate the database analysis and processing capabilities, create the database and the 25 TPC-DS built-in shardings in advance.

Step 3: Parameters Tuning

 Tune relevant parameters as per the suggestions of GBase.

Step 4: Data Generation

 TCH-H can generate required test data by setting parameters through the dbgen tool. The command parameter is /dbgen -C 10 -S 1 -s 3000 –vf;

 TPC-DS can generate required test data by setting parameters through the dsdgen tool. The command parameter is dsdgen -scale 3000 -dir testdata -force -parallel 10 -child 1.

Step 5: Data Loading

 When importing the data, an FTP server is built on the local node of the GBASE cluster and data is loaded through FTP. When a sharding contains multiple data files, the multiple data files of a single sharding are integrated into a single import sentence, and the test data is separately loaded into 8 database shardings at the granularity of a single sharding.

Step 6: SQL Execution

 Independent client communicates with GBase Cluster through service plane network, and uses gccli tool to execute TPC-H 22 SQL use cases, reference command is as follows:

/home/GBase/gccli_install/gcluster/server/bin/gccli -h 10.28.100.38 -uroot -Dtpch -vvv < query_1.sql.

3 Verification Results

The Verification Results  under GBase 8a MPP scenario are as follows:

3.1 Performance of Union Memory SAS SSD under Raid Logical Volume Bandwidth Performance Testing

Fig. 3: Performance of Union Memory SAS SSD under Raid Logical Volume Bandwidth Performance Testing

In this test, 6 SAS SSDs and 12 SATA SSDs are configured with Raid5 and Raid50, respectively. The FIO test tool is used for pressure testing of 128KB sequential read/write bandwidth performance on the Server host. As shown in Fig. 3, SAS SSDs have better read/write bandwidth. The read bandwidth is increased by about 19.6%  and the write bandwidth is increased by about 34%  compared with SATA, showing the absolute advantage of SAS SSDs under the Raid logical volume bandwidth.

3.2 Performance of Union Memory SAS SSD under Degraded Single Disk at Fault & Reconfigured Write Performance

Fig. 4: Performance of Union Memory SAS SSD under Degraded Single Disk at Fault & Reconfigured Write Performance

The Read Performance of Degraded Single Disk at Fault refers to the read service performance of the logical volume of the RAID group when a single disk is at fault or plugged out. The Reconfigured Performance refers to the service-side performance when the RAID group simultaneously performs data reconfiguration and service delivery on the hot spare disk when a single disk is at fault. This test is performed when 6 Union Memory SAS SSDs are configured with Raid5 and one SSD fails.

As shown in Fig. 4, SAS SSDs perform better than SATA SSDs both in the read performance of degraded single disk at fault and in the reconfigured read bandwidth. The read performance of degraded single disk at fault is about 22.7%  better than that of SATA SSDs, and the reconfigured read bandwidth performance is about 38.4%  higher than that of SATA SSDs.

3.3 Performance of Union Memory SAS SSD under TPC-DS Scenario

Fig. 5: Performance of Union Memory SAS SSD under TPC-DS Scenario

TPC-DS test is to load data to GBase database shardings via FTP protocol. As can be seen from Fig. 5, under the same physical hardware environment, Union Memory SAS SSDs take slightly less time than SATA SSDs in both data import and SQL use case execution, and the former have a certain advantage in time.

3.4 Performance of Union Memory SAS SSD under TCP-H Scenario

Fig. 6: Performance of Union Memory SAS SSD under TCP-H Scenario

TPC-H test is to load data to GBase database shardings via FTP protocol. Fig. 6 shows the performance of Union Memory SAS SSDs and SATA SSDs under TPC-H test. The total time of Union Memory SAS SSDs is slightly less than that of SATA SSDs, the data import time of Union Memory SAS SSDs is about 6%  less than that of SATA SSDs, and the execution time of SQL use case with Union Memory SAS SSDs is reduced about 3%  than that of SATA SSDs.