I prefer to have a general simple calculation for health, rather than multiple tiers of warnings. The other thought I had for warning is: (busy/max) > 75% peter lin
Antonio_Fiol_Bonn�n <[EMAIL PROTECTED]> wrote: Hi, I would establish two thresholds (possibly in %). Or maybe one threshold could be "spare" and the other "spare/4" or something. max == busy seems too high for a simple "warning". Apache web server prints an "error" (not a warning IIRC) on the logs if MaxClients is reached. Or maybe I dod not unserstand your proposal... Antonio Fiol Peter Lin wrote: > >Here is a bit more information on how I am implementing the monitor. to calculate the >health I am using max, busy and spare threads. > >healthy = ( max - busy) > spare >active = (max - busy) < spare >warning = (max == busy) > > >Do people think that is reasonable? > >peter lin > > >Peter Lin wrote: > >Some people already know this, but most don't. I am working on a monitor for Tomcat >using JMeter. The first graph will be the status. I made a mockup in photoshop. The >actual UI hasn't been coded yet. I'd like to hear people's thoughts and comments. > >http://cvs.apache.org/~woolfel/server-monitors-status-mockup.jpg >Here is how I am calculating the load. It is a combination of memory and thread. > >memory load = (used memory/ max memory) * load factor >the default load factor is 33 > >thread load = (busy threads/ max threads) * load factor >the default load factor is 67 > >tomcat load = memory load + thread load > >I plan to store the load factor in JMeter's properties file, so users can change the >value. There are numerous limitation to this approach that I can think of, but I'm >sure others can think of other limitations. > >1. a small number of threads could be taking 100% of the CPU, like XML. which would >mean both memory and thread don't tell you accurate what the load is > >2. a large number of concurrent requests for static files may not place load on the >system > >3. memory usage could be high, but the system may be responding within acceptable >ranges > >The load calculation at this point is at a high level, but I feel it is still useful. >Ultimately the load calculation will have to be specific to the actual webapps. Since >JMeter and the new monitor will be extensible, users should be able to provide their >own class to calculate load. Once I get a working version done. I can't say when it >will get done, but I am actively working on it. > >peter lin > > > >--------------------------------- >Do you Yahoo!? >Protect your identity with Yahoo! Mail AddressGuard > >--------------------------------- >Do you Yahoo!? >Yahoo! Search - Find what you�re looking for faster. > > > ATTACHMENT part 2 application/x-pkcs7-signature name=smime.p7s --------------------------------- Do you Yahoo!? Yahoo! Search - Find what you�re looking for faster.
