Buffer cache structure

Sybase IQ automatically calculates the number of cache partitions for the buffer cache according to the number of CPUs on your system. If load or query performance in a multi-CPU configuration is slower than expected, you may be able to improve it by changing the value of the CACHE_PARTITIONS database option. For details, see CACHE_PARTITIONS option in Sybase IQ Reference Manual.

As buffers approach the Least Recently Used (LRU) end of the cache, they pass a wash marker. Sybase IQ writes the oldest pages—those past the wash marker—out to disk so that the cache space they occupy can be reused. A team of Sybase IQ processing threads, called sweeper threads, sweeps (writes) out the oldest buffers.

When Sybase IQ needs to read a page of data into the cache, it grabs the LRU buffer. If the buffer is still “dirty” (modified) it must first be written to disk. The Gdirty column in the monitor -cache report shows the number of times the LRU buffer was grabbed dirty and Sybase IQ had to write it out before using it.

Usually Sybase IQ is able to keep the Gdirty value at 0. If this value is greater than 0 for more than brief periods, you may need to adjust one of the database options that control the number of sweeper threads and the wash marker. See “SWEEPER_THREADS_PERCENT option” or “WASH_AREA_BUFFERS_PERCENT option” in Chapter 2, “Database Options,” Sybase IQ Reference Manual.