huh, I was wondering why this patch does not receive any feedback from advocates.... So, I found out, I haven't done RTI build and therefore also haven't mailed the request... sigh.

Any dtrace experts around to advise me? I'm looking into a performance issue where I'm measuring the run time of a function with dtrace FBT entry/return probes, collecting an aggregation of run times base on return timestamp minus entry timestamp.

The resulting distribution is bi-modal, with the majority of samples in an expected range, and a smaller group taking much, much longer than the majority group.

What's a good way to get details on the calls below this function, just for the minority cases that take longer?

You might try speculative tracing

gwr: ^^

thanks

[illumos-gate] 15089 WPTS BVT_TreeMgmt_SMB311_Disconnect_NoSignedNoEncryptedTreeConnect -- Gordon Ross <gwr⊙rc>