"Discovered" this guy due to some reddit post. Now reading his "Systems Performance" book which should definitely be required reading for performance engineers.
This is a summary / cheat-sheet for diagnosing performance problems with the "USE" method. It's explained in much more detail here: http://www.brendangregg.com/usemethod.html
Worth checking out the rest of the site too, really useful stuff for performance engineers.
For reference, the definitions are:
* utilization: the average time that the resource was busy servicing work
* saturation: the degree to which the resource has extra work which it can't service, often queued
He described saturation as how much the processes want to use the resource beyond the maximum 100% utilization. E.g. if you look at load average on Linux, that's the size of the run queue (processes that are ready to run + blocking on uninterruptable IO) versus just CPU usage which cannot go over the 100%. That requires there is a meaningful queue. There's lots more in his excellent "Systems Performance" book.
Comments
"Discovered" this guy due to some reddit post. Now reading his "Systems Performance" book which should definitely be required reading for performance engineers.
A good collection of tools, but I can't figure out the distinction between utilization and saturation in this schema.
This is a summary / cheat-sheet for diagnosing performance problems with the "USE" method. It's explained in much more detail here: http://www.brendangregg.com/usemethod.html
Worth checking out the rest of the site too, really useful stuff for performance engineers.
For reference, the definitions are:
* utilization: the average time that the resource was busy servicing work
* saturation: the degree to which the resource has extra work which it can't service, often queued
* errors: the count of error events
He described saturation as how much the processes want to use the resource beyond the maximum 100% utilization. E.g. if you look at load average on Linux, that's the size of the run queue (processes that are ready to run + blocking on uninterruptable IO) versus just CPU usage which cannot go over the 100%. That requires there is a meaningful queue. There's lots more in his excellent "Systems Performance" book.
re: storage capacity, does df -h report inodes on other OS? On Ubuntu you need to do df -i to see inode utilization.
it is like TL:DR for "Systems Performance" book :)