 |
 |
Hi Catalin, Jacek has the actual figures on what we had to cope with monitoring i/o events on 30 servers running flat out. If I remember correctly, the packet rate was sufficiently high that the collector had little time left to do any kind of meaningful analysis and was spending most of it's time logging the data. Jacek can you provide a better assesment here? Andy On Thu, 28 Apr 2005, Catalin Cirstoiu wrote: > Hi Ady, > > We developed a small plugin called ApMon for Application Monitoring. > This also sends UDP datagrams containing multiple parameters from > instrumented applications to a MonALISA service. We managed to send and > process on the fly about 5000 datagrams per second without any packet > loss. If, for example, each node would send one datagram each 30s, I > think that this processing could be easily done in real time. > > Cheers, > Catalin. > > Andy Hanushevsky wrote: > > Hi Catlin, > > > > Just to make sure we are on the same "page" so to speak. You can get > > auto-flushing but you must create a monitoring stream that does not > > include I/O events. Monitoring streams that include I/O events cannot be > > flushed. So, if you add in files events then you wind up delaying files > > information as well. If you really need i/o events (generally, these > > events tell you little about the server and are only useful for tuning > > an application), send them to an offline data collector. Beleive me, in > > a real-life system you will not be able to analyze i/o events in real time. > > > > Andy > > >