196 POWER7 and POWER7+ Optimization and Tuning Guide
LDR_CNTRL=TEXTPSIZE=64K@STACKPSIZE=64K@DATAPSIZE=64K (Firm suggestion)
Set in the
Sybase owner profile, not in start_asiq, this environment variable enables the
application to use 64 KB pages.
AIXTHREAD_PREALLOC=<
size of IQ heap in bytes> (Firm suggestion)
– A mostly undocumented feature of AIX 5L V5.3.0.10+ and AIX V6.1.0.3+
– The kernel pre-sets the heap pointer, reducing the number of system calls in malloc
– Saves some time that is otherwise taken by threads that are waiting on the
malloc mutex
– Reported to benefit larger systems, 8-core or more
– Probably most beneficial to IQ load phases
Special considerations for Sybase IQ with SMT4 mode
SMT4 mode provides special benefits for Sybase IQ, owing to its highly threaded
architecture. The IQ configuration parameter, -iqnumbercpus, is determined by default by the
number of logical processors that IQ detects. The -iqnumbercpus parameter affects the value
of several downstream parameters, including -iqgovern and -iqmt. The configuration
parameter -iqmt governs the number of threads created.
Some special considerations are:
By default, -iqmt is set to 60 * (min(-iqnumbercpus,4)) + 50 * (-iqnumbercpus - 4) +
2*(numConnections + 2) + 1
On a 4-core, single threaded system with -gm (number of connections), it is set to 20, that
is, iqmt = 60 * (4) + 50 * (4-4) + 2 * (20+2) + 1 =285
However, if SMT4 mode is on, -iqnumbercpus is set to 16 for a 4-core system, so -iqmt is
1605. Therefore, 1605 threads are allocated by IQ.
Is IQ allocating too many threads in SMT4 mode? With POWER5 SMT2 mode, it is
beneficial to set -iqnumbercpus to the number of cores or virtual processors. However,
with POWER6 and POWER7, the best performance is generally obtained by leaving
-iqnumbercpus at its default, even though IQ creates more threads.
Shared versus dedicated processors with Sybase IQ and POWER7
Sybase IQ is CPU hungry by design and scale both out and up with the addition of
processors and workload. Sybase support generally favors dedicated cores over shared
processor pool technology, and some enterprises report some negative experiences with IQ
on shared processors, which might be because of extraneous factors.
The IBM team conducted systematic comparisons of dedicated and shared processor
performance using a Sybase provided test workload to produce the Stockmarket Benchmark.
This benchmark is similar to TPC-H in that it consists of simulated data that can be inflated.
There are 18 complex queries that can be run in parallel or one at a time.
To Next Page
Loading...
