The information given in “Sizing the SQT cache,” in Chapter 16, “Performance Tuning,” in the Replication Server Administration Guide Volume 2 is updated due to changes in Replication Server version 12.6.
Replication Server 12.6 introduced the sqt_init_read_delay and sqt_max_read_delay configuration parameters to address SQT performance issues. These configuration parameters and their effects come into play regardless of the setting of SMP. If you observe a slow-down in performance after upgrading from Replication Server 12.5 to Replication Server 12.6, these configuration parameters should be set to their minimum values.
Increasing sqt_max_cache_size was used for tuning in Replication Server 12.5. In Replication Server 12.6 the benefit of SQT caching remains but oversizing it can be counter-productive. If an installation has a very large SQT cache, any slowdown in performance from Replication Server 12.5 to Replication Server12.6 could be attributed to oversizing the SQT cache. The problem can be avoided by decreasing the setting for sqt_max_cache_size. SQT cache is sufficiently sized if you do not observe, or observer only infrequently, transactions being flushed from cache because there is not enough room to store more in cache.
The Data Server Interface (DSI) may have problems in the handling of large transactions. Even in the case when single DSI is being used, instead of parallel DSI, the identification of large transactions by the DSI can spell performance difficulties. The solution is to configure DSI so that it does not recognize any large transactions. You can do this by setting dsi_large_xact_size to its maximum value of 4,294,967,295.
