Ramon,
Thanks for that piece of information. It will help explain why there is never a blip on that graph. After some more investigation, I found that the processes flagged "Nice" are the reason behind the values not adding up to near 100%. Ganglia 3.0.1 does not appear to use the "Nice" CPU statistics in any of the calculations or totals. I'll probably have to ask this in the dev forum, but is this working as intended? If you are not going to differentiate between user and nice, shouldn't you at the very least lump them together? *I hope the last one isn't a holy war question. If so, please forgive* Regards, Bart Hampshire _____ From: Ramon Bastiaans [mailto:[EMAIL PROTECTED] Sent: Wednesday, October 12, 2005 1:47 AM To: Bart Hampshire Cc: [email protected] Subject: Re: [Ganglia-general] CPU Metrics Question In addition to previous posts; Currently - in version 3.0.1 of Ganglia - the cpu_wio metric is broken and not reported correctly. This is the amount of time the cpu spends waiting for I/O. This could account for the remaining percentage of cpu usage that you are missing. In the upcoming 3.0.2 release cpu_wio metric should be fixed. Kind regards, - Ramon. Bart Hampshire wrote: Mike, Thanks for taking a stab at it. Double checking the data, 0.64 is really 0.64% and not 64%. ps on the box doesn't indicate that gmond has a nice value of 0. However, it does show that some of the user processes we'd expect on the box have nice values of 15 and not 0. These processes are running at nice level of 0 on other machines. Could this be a cause of the behavior? Maybe if I rephrase the question it might help... Can anyone think of a reason that the 4 cpu metrics (user+system+wio+idle) would not sum to near 100%? On a dual CPU system? Regards, Bart Hampshire -----Original Message----- From: [EMAIL PROTECTED] [mailto:ganglia-general- [EMAIL PROTECTED] On Behalf Of michael chang Sent: Saturday, October 08, 2005 6:41 PM To: Bart Hampshire Cc: Subject: Re: [Ganglia-general] CPU Metrics Question On 10/8/05, Bart Hampshire <mailto:[EMAIL PROTECTED]> <[EMAIL PROTECTED]> wrote: Hello all, Was wondering if someone might be able to explain why it is that we often see total CPU Utilization (User + System + WIO + Idle) at values substantially below 100%. When this happens, the reported user portion is almost always zero while running top on the box shows that 90+% of the CPU(s) is(are) being utilized by user processes. The XML stream on the gmond collector (we are using gmond in unicast mode) node is reporting the values that appear recorded in the gmetad's RRDs. From this, it appears that the CPU User metric isn't registering the utilization properly. Linux kernel versions are 2.4.x and 2.6.x on Red Hat and SLES x86 boxes. Sample data collected from RRD's to follow sig in which the reported total sum of CPU Utilization was below 10%. 7-9-2005 13:10 User: 0.00 System: 0.56 WIO: 0.00 Idle: 0.09 Total: 0.64 7-9-2005 13:15 User: 0.00 System: 0.60 WIO: 0.00 Idle: 0.10 Total: 0.70 7-9-2005 13:20 User: 0.00 System: 0.60 WIO: 0.00 Idle: 0.10 Total: 0.70 7-9-2005 13:05 User: 0.00 System: 1.96 WIO: 0.00 Idle: 0.06 Total: 2.02 7-9-2005 13:10 User: 0.00 System: 1.30 WIO: 0.00 Idle: 0.00 Total: 1.30 7-9-2005 13:15 User: 0.00 System: 1.30 WIO: 0.00 Idle: 0.00 Total: 1.30 7-9-2005 13:20 User: 0.04 System: 1.65 WIO: 0.00 Idle: 0.90 Total: 2.59 It looks like 0.64 represents 64%, not 0.64 percent; or whatever types of numbers would be reported by a program such as tload. That said, maybe ganglia is running at too high or too low of a nice/priority value. Or your CPU has been downclocked or something else funky. -- ~Mike - Just my two cents - No man is an island, and no man is unable. ------------------------------------------------------- This SF.Net email is sponsored by: Power Architecture Resource Center: Free content, downloads, discussions, and more. http://solutions.newsforge.com/ibmarch.tmpl _______________________________________________ Ganglia-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-general ------------------------------------------------------- This SF.Net email is sponsored by: Power Architecture Resource Center: Free content, downloads, discussions, and more. http://solutions.newsforge.com/ibmarch.tmpl _______________________________________________ Ganglia-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-general -- .--------------------------------------------. | ing. Ramon Bastiaans | | HPC - Systems Programmer | |--------------------------------------------| | SARA - Computing and Networking Services | | Kruislaan 415 PO Box 194613 | | 1098 SJ Amsterdam 1090 GP Amsterdam | |--------------------------------------------| | Mail: bastiaans ( a t ) sara ( d o t ) nl | | Web: http://www.sara.nl/ | | Phone: +31 (0)20 592 80 19 | | Fax: +31 (0)20 668 31 67 | `--------------------------------------------'
