SS performance issue with pruning connections during discovery

book

Article ID: 76554

calendar_today

Updated On:

Products

CA Spectrum

Issue/Introduction

When running a discovery of HP BG devices, the SS cpu spiked to 100% and OneClick became extremely slow.  When running pstacks on the SpectroSERVER process, the SS is busy in this action:  CsPort::prune_pconns_by_proximity

Thread 1 (Thread 0x7fffec5e8720 (LWP 11211)):
#0  0x00007ffff7282c30 in CsNode::cmp_and_set_local_to_global () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#1  0x00007ffff72cfe08 in CsPort::check_device(CsPort&) () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#2  0x00007ffff72cff00 in CsPort::check_port(CsPort&) () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#3  0x00007ffff72cfe17 in CsPort::check_device(CsPort&) () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#4  0x00007ffff72d0010 in CsPort::check_pconn(CsPort&, CsPort&, CsPorts const*) () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#5  0x00007ffff72d07a1 in CsPort::prune_pconns_by_proximity() () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#6  0x00007ffff72c6deb in prune_pconns_wrapper(void*) () from /opt/SPECTRUM/lib/../SS/libadcor.so.1
#7  0x00007ffff7762a63 in moot_thread_start () from /opt/SPECTRUM/lib/libmoot.so.1
#8  0x0000003ac2043760 in ?? () from /lib64/libc.so.6
#9  0x0000000000000000 in ?? ()

Cause

The problem is due to the thread priority for pruning connections was higher than the normal SpectroSERVER processing threads.  The thread priority needed to be changed to a lower priority to allow the SS to OC threads to continue to take priority.

Environment

Release:
Component: SPCCSS

Resolution

This is tentatively scheduled to be fixed in CA Spectrum 10.3 once released.