I'd like to remind all of you that when I ported the monitoring core to
Solaris, I never tested it on more than a four-way E420R (or on anything
pre-Solaris 7) ... I tested it on the widest range of Sun iron available
to me, which isn't very much.
This looks like a Steve's Crappy Coding problem. I probably did not
think too much about the array that holds CPU cycle counters for each
CPU, and my bad coding is manifesting when that data structure grows to
a certain point. Sorry about that.
I noticed two problems during testing - one, for some reason kvm_read()s
would fail intermittently and I couldn't track down exactly why. And
two, the CPU tallies never felt quite right, but they were delivering
numbers as accurate as top's.
Unfortunately I don't have much time to devote to Ganglia these days
(now that everything's running smoothly at my installation), so unless
someone else wants to step up and debug/fix this, I don't know what
realistic solution I could offer you...
Ng Hee Khiang wrote:
The following is the debug output from running gmond on solaris 2.6
gmond start -d 99
pthread_attr_init
creating cluster hash for 1024 nodes
hash_create size = 1024
hash->size is 1031
gmond initialized cluster hash
Using interface hme0
mcast listening on 239.2.11.71 8699
XML listening on port 8649
listening thread(s) have been started
mcast_listen_thread() started 5
listening thread(s) have been started
cleanup thread has been started
multicasting on channel 239.2.11.71 8699
Lookup up kstat: km (unix?)='unix', ks (system_misc?)='system_misc',kn
(resulting metric?)='ncpus'
ncpus: kc is 381f8
ncpus: Just did kstat_lookup().
ncpus: Looked up.
ncpus: Kstat data type: 2, Value returned: 6, 6 6 6
ncpus: Kernel close. Val returned: 6
metric_init: Assigning cpu_num value (6) to ncpus.
running update_metric_data() ...
Lookup up kstat: km (unix?)='unix', ks (system_misc?)='system_misc',kn
(resulting metric?)='boot_time'
boot_time: kc is 381f8
boot_time: Just did kstat_lookup().
boot_time: Looked up.
boot_time: Kstat data type: 2, Value returned: 1047269674, 1047269674
1047269674 1047269674
boot_time: Kernel close. Val returned: 1047269674
Lookup up kstat: km (unix?)='cpu_info', ks (system_misc?)='cpu_info0',kn
(resulting metric?)='clock_MHz'
clock_MHz: kc is 381f8
clock_MHz: Just did kstat_lookup().
ks: Not enough space
clock_MHz: Looked up.
Segmentation Fault
-----Original Message-----
From: Ken MacInnis [mailto:[EMAIL PROTECTED]
Sent: Wednesday, May 07, 2003 8:39 PM
To: Ng Hee Khiang
Cc: [email protected]
Subject: Re: [Ganglia-general] Gmond on Solaris 2.6/2.8
On Wed, 7 May 2003, Ng Hee Khiang wrote:
I had installed gmond on solaris machine, with OS 2.6 - 2.8 ,
however , when i try to run gmond , it does not start.
Any suggestion to compile ganglia for solaris ??
Use a 64-bit compiler. GCC 3.2.2 configured as 'sparcv9-sun-solaris2.9'
worked fine for me. Check the debugging output at start-up with the '-d
99' flag, you should get output warning you of this if you didn't.
