Menu
▾
▴
Re: [Collectl-interest] Suggestion: Additional NFS data from /proc/self/mountstats (same as nfsiostat-command)
|
From: Thomas O. <tho...@er...> - 2016-03-14 15:45:13
|
Hi Mark, Thanks! Yes, another switch.. How I have waited! ☺ I looked at /proc/self/mountstats and could not see any timingdata either.. So where nfsiostat gets the RTT values is a bit of a mystery. I did find some references of a nfs-iostat.py script that might give a clue. http://git.linux-nfs.org/?p=steved/nfs-utils.git;a=blob;f=tools/nfs-iostat/nfs-iostat.py;h=9626d42609b9485c7fda0c9ef69d698f9fa929fd;hb=HEAD) I think it runs several times and calculates delta?! If it helps, this is output from nfsiostat on one of our RedHat 6.5 servers: (As you can see the RTT for write operations are very high). [root@myserver collectl]# nfsiostat 10 2 /proj/eiffel002_config_fem001 nfsserv.somedomain.se:/vol/volp01234/data_config mounted on /proj/data_config_server001: op/s rpc bklog 218.81 0.00 read: ops/s kB/s kB/op retrans avg RTT (ms) avg exe (ms) 12.293 697.798 56.763 0 (0.0%) 7.972 15.149 write: ops/s kB/s kB/op retrans avg RTT (ms) avg exe (ms) 38.325 1322.385 34.505 392 (0.0%) 35.313 1237.445 nfsserv.somedomain.se:/vol/volp01234/data_config mounted on /proj/data_config_server001: op/s rpc bklog 68.20 0.00 read: ops/s kB/s kB/op retrans avg RTT (ms) avg exe (ms) 0.000 0.000 0.000 0 (0.0%) 0.000 0.000 write: ops/s kB/s kB/op retrans avg RTT (ms) avg exe (ms) 69.700 4481.492 64.297 0 (0.0%) 658.139 71380.063 I understand that this is hard to build and test without NFS systems! I just wanted to throw out the suggestion and see what happens. Learning to write a plugin for collectl is tempting. I am not a programmer, but have fiddled around with some simple perl scripts in the past. I’ll do some reading on the webpage and try to get an idea of how the plugin stuff works. You’ll never get rid of me if I start.. ☺ Kind Regards, Thomas From: Mark Seger [mailto:mj...@gm...] Sent: den 14 mars 2016 14:56 To: Thomas Oliw Cc: col...@li... Subject: Re: [Collectl-interest] Suggestion: Additional NFS data from /proc/self/mountstats (same as nfsiostat-command) Always happy to hear from happy users. I just looked at /proc/xx/mountstats, which actually applies to all pids, self is just a shortcut to yourself. The problem with pid-based stats is it can be a lot of overhead to read any more stats than collectl already reads, but my thought was I might be able to add something optionally. Oh boy, another switch! ;) But when I looks at these stats I did't see anything about timing and only saw info on what is mounted. That said, I'd think since nfs is a shared resource, there might be timing data for nfs in generat, but my systems currently don't use nfs and I might need to do some experiments to see what happens if/when I do configure it. Worse case, especially if you're a collectl fan, you might be able to write your own plugin if you're a perl user. The benefit there is once you see how easy it is to write a plugin you then might be able to add even more metrics, possibly at the application level if you find that useful. If so, I'm always ready to help... I'm out of town this week but I'll try to revisit next week when I return. -mark On Mon, Mar 14, 2016 at 8:12 AM, Thomas Oliw <tho...@er...<mailto:tho...@er...>> wrote: Hi, I love collectl and use it extensively for many performance related troubleshooting/monitoring tasks in our server park. The possibility to run live and/or record to file is a fantastic mix of features and very useful! However, one thing that I miss, is NFS Response time data… We use lots of NFS shares in our environment, and that particular metric is one of the most useful ones in my opinion. As a complement to collectl, I use “nfsiostat” when NFS is suspected to be a performance bottleneck. It shows me a number of good metrics and has a “RTT” (Round Trip Time) field, that at least gives me a hint of the NFS server responsetime. If I read the documentation correct, it gets its data from /proc/self/mountstats. I think it would be very useful if those metrics could be collected in collectl as well. The nfsiostat tool itself is a bit crude, at least in our a bit aged RedHat environment and for us it would be convenient to have these metrics managed with collectl instead. Just a suggestion… Thanks for the collectl tool!! Kind Regards, Thomas Oliw ------------------------------------------------------------------------------ Transform Data into Opportunity. Accelerate data analysis in your applications with Intel Data Analytics Acceleration Library. Click to learn more. http://pubads.g.doubleclick.net/gampad/clk?id=278785231&iu=/4140 _______________________________________________ Collectl-interest mailing list Col...@li...<mailto:Col...@li...> https://lists.sourceforge.net/lists/listinfo/collectl-interest |