The graphical plan feature in Interactive SQL displays the execution plan for a query. It is useful for diagnosing performance issues with specific queries. For example, the information in the plan can help you decide where to add an index to your database. You can save the graphical plan for a query for future reference by choosing File > Save Plan in Interactive SQL. SQL Anywhere graphical plans are saved with the extension .saplan.
Graphical plans are also available via Application Profiling mode in Sybase Central. For more information about the Application Profiling features of Sybase Central, see Application profiling.
The graphical plan provides a great deal more information than the short or long plans. You can choose to see the graphical plan either with or without statistics. Both allow you to view quickly which parts of the plan have been estimated as the most expensive. The graphical plan with statistics, though more expensive to view, also provides the actual query execution statistics as monitored by the database server when the query is executed, and permits direct comparison between the estimates used by the optimizer in constructing the access plan with the actual statistics monitored during execution. Note, however, that the optimizer is often unable to precisely estimate a query's cost, so expect there to be differences. The graphical plan is the default format for access plans.
You can obtain detailed information about the nodes in the plan by clicking the node in the graphical diagram. The graphical plan with statistics shows you all the estimates that are provided with the graphical plan, but also shows actual runtime costs of executing the statement. To do this, the statement must actually be executed. This means that there may be a delay in accessing the plan for expensive queries. It also means that any parts of your query such as deletes or updates are actually executed, although you can perform a rollback to undo these changes.
Use the graphical plan with statistics when you are having performance problems, and the estimated row count or run time differs from your expectations. The graphical plan with statistics provides estimates and actual statistics for you to compare. A large difference between actual and estimate is a warning sign that the optimizer might not have sufficient information to prepare correct estimates.
Following are some of the key statistics you can check in the graphical plan with statistics, and some possible remedies:
Row count measures the rows in the result set. If the estimated row count is significantly different from the actual row count, the selectivity of underlying predicates is probably incorrect.
Accurate selectivity estimates are critical for the proper operation of the optimizer. For example, if the optimizer mistakenly estimates a predicate to be highly selective (with, say, a selectivity of 5%), but in reality, the predicate is much less selective (for example, 50%), then performance may suffer. In general, estimates may not be precise. However, a significantly large error does indicate a possible problem. If the predicate is over a base column for which there does not exist a histogram, executing a CREATE STATISTICS statement to create a histogram may correct the problem. If selectivity error remains a problem then, as a last resort, you may want to consider specifying a user estimate of selectivity along with the predicate in the query text.
Runtime measures the time to execute the query. If the runtime is incorrect for a table scan or index scan, you may improve performance by executing the REORGANIZE TABLE statement. You can use the sa_table_fragmentation and the sa_index_density system procedures to determine whether the table or index are fragmented.
When the source of the estimate is Guess, the optimizer has no information to use, which may indicate a problem. If the estimate source is Index and the selectivity estimate is incorrect, your problem may be that the index is skewed: you may benefit from defragmenting the index with the REORGANIZE INDEX statement.
If the number of cache reads and cache hits are exactly the same, then your entire database is in cache—an excellent thing. When reads are greater than hits, it means that the database server is attempting to go to cache but failing, and that it must read from disk. In some cases, such as hash joins, this is expected. In other cases, such as nested loops joins, a poor cache-hit ratio may indicate a performance problem, and you may benefit from increasing your cache size.