Size of database copies

To evaluate the size of the database copies influence on the transaction processing with the usage of Theta replication technique, experiments with different sizes of replicas are conducted. Again, the load in experimental scenarios changes from 5 up to 100 tps and 100% of submitted transactions are updates.

Name Values

Data management Theta replication method Replicas 2

Size of tables 10K, 100K, 1M, 10M, 100M [rows]

Load 5-100 [tps]

Queries percentage 0 [%]

Conflicts ratio 20 [%]

Table 4.10: Scenario with table indexes – transient transactions

The tests are performed for varying sizes of database tables with indexes, which simulates transient transactions (scenario in table 4.10), and without key indexes – long lasting transactions (table4.12). Test results are presented respec-tively in tables 4.11 and 4.13.

Removal of table indexes in the scenario without table indexes causes that full table scans need to be performed on these tables while processing transactions.

This slightly increases the duration of their processing in the database, and in a consequence queries run against non-indexed tables appropriately simulates long lasting queries. Therefore, results gained in these 2 scenarios show how the system is affected when the type of the load changes.

In the scenarios where table indexes exists in database tables, the size of tables in each database copy changes from 10 thousand rows up to 100 million rows.

Since database response times for scenarios without indexes increased consider-ably, tests for this scenario are performed only up to 1 million rows.

Figure 4.7: Database size influence on mean response times – tables with indexes

Mean response times for replicated system with key indexes existing are pre-sented in fig. 4.7. It is noticeable that response times grow slightly for the lower loads. For the loads higher than 50 tps the average database response time grows around 10 times (from 3 ms for 10K rows to 29 ms for 100M rows) and mean re-sponse time increases faster. However, this growth is fairly slow with a reference

to the growth of the database response time, and the system continuously replies within acceptable limits.

Load [tps] 10K 100K 1000K 10M 100M

5 57.0 58.0 58.0 57.2 60.2

70 93.8 93.6 94.6 109.6 111.4

80 98.8 101.0 106.4 113.8 119.4 90 106.6 119.6 118.0 119.2 127.2 100 116.6 119.6 121.8 128.8 137.2

Table 4.11: Test results for transient transactions

Scenario without table indexes, which simulates long lasting transactions, is presented in table 4.12. Mean response times for this scenario are graphically presented in fig. 4.8, while tests results are supplied in table 4.13.

Name Values

Data management Theta replication method Replicas 2

Size of tables 10K, 100K, 1M [rows]

Load 5-100 [tps]

Queries percentage 0 [%]

Conflicts ratio 20 [%]

Table 4.12: Scenario without table indexes – long lasting transaction

In the scenario when there are no key indexes in tables, the mean database response time grows rapidly and for 1 million rows (mean database response time is 59 ms) this time is more than 10 times longer than for the non-indexed tables.

Such a high growth in the time of a single long transaction processing in database causes that system performs efficiently only for the lower loads. For the

higher loads in connection with tables containing 1 million rows mean response time grows very quickly, which may cause serious bottlenecks in a system.

The experiment results presented in the scenarios in this section prove that Theta method allows to gain very good mean response times for large amount of short transactions.

On the other hand, when the database response time grows significantly, the approach is not suitable when it comes to the higher loads. However, the method can still be applicable for synchronization of such database copies when the load is lower.

Figure 4.8: Database size influence on mean response times – long lasting trans-actions for tables without indexes

Since the approach is destined to transactional, multi-node systems, issue related to high loads for long lasting transactions is not further analyzed.

Load [tps] 10K 100K 1000K

Table 4.13: Test results for long lasting transactions

4.3.5 Portability

To perform portability tests of Theta replication approach, replication environ-ment is prepared in configuration with several hardware and software platforms.

On the purpose of this experiment Theta method is implemented for the SUN SUN SPARC, AMD and Intel platforms, SUN Solaris and RedHat operating sys-tems and Oracle and PostgreSQL database engines.

The performed tests in various configurations confirm Theta replication method portability and possibility of the usage in heterogeneous environments based on different operating systems, various hardware platforms and RDBMS.

4.4 Summary

Similarly to the evaluation methods of the replication approaches available in literature, Theta replication method evaluation is performed in a dedicated test environment, which allows to verify whether it is applicable for the certain solu-tions that it is designed and implemented for.

The chapter covers the issues related to the experimental evaluation of new Theta replication method in aspects of its scalability and portability. Presented tests results and their analysis show that Theta replication provides scalable and

non-conflicting mechanism that allows on keeping multiple replicas synchronized.

The method ensures high scalability and at the same time provides acceptable efficiency of data replication process. Theta method ensures complete data con-sistency and its performance is very good comparing to a centralized system for all testing workloads.

• Performed experiments show that the system still responses efficiently while the amount of replicas is growing and for higher transaction loads replicated system behaves better than a single instance of centralized database.

• In-laboratory method implementation of the system with Theta replica-tion is compatible with the most of existing database systems. Since the approach is based on the execution of the stored procedures in remote repli-cas, it can be used in heterogeneous environments with no difficulty.

• Conducted experiments confirm Theta replication method portability and possibility of the usage in environments based on different operating sys-tems, various hardware platforms and RDBMS.