Possible bad config node: Eval.JitWarmupRequests

[JohnGW05]

HHVM Version

HipHop VM 3.16.0-dev (rel)
Compiler: heads/master-0-g8920e2cd8f9251978713612dcc6891d7334a5d96
Repo schema: 922cef7ce4ceb4cc7b7ff58c0cc0972ee69ae429

Standalone code, or other way to reproduce the problem

i set following in the config file:
Jit = true
JitWarmupRequests = 0

Expected result

hhvm run as fastcgi proxygen and jit enabled at begin

Actual result

root@localhost:~/gw05/nginx# /usr/local/bin/hhvm -c /usr/local/etc/hhvm/hhvm.hdf -m server -d hhvm.jit_warmup_requests = 0 -vEval.JitProfileInterpRequests=0 -d hhvm.jit_profile_interp_requests=0 -vhhvm.jit_profile_interp_requests=0 -v Eval.JitProfileInterpRequests=0 -v Eval.Jit=1 -v Eval.JitProfileInterpRequests=0
Possible bad config node: Eval.JitWarmupRequests
Possible bad config node: Eval.JitAStubsSize
Possible bad config node: Log.InjectedStackTrace
Possible bad config node: Server.DanglingWait

question:
so, 1、why hhvm didn't warmup jit at the fisrt ?how to make the jitWarmupRequests=0works?
2、i run "webbench -c 200 -t 60 http://10.130.48.73:80/bench.php" in the other client machine,when i set jit = true or jit = false, the results is the same, so does the jit called? and how can i know if the jit works?
thank you

[paulbiss]

Well hhvm.jit_warmup_requests isn't an hhvm option. The first Eval.JitProfileInterpRequests run in the interpreter and it looks like you set that to zero already.

You need to send enough requests to the jit for it to (1) generate profile translations, (2) collect profile data, and (3) generate optimized translations. The hhvm JIT is tuned based on profile data, and running without these optimizations would be highly inadvisable. At a minimum the first set of requests will need to actually generate the machine code which can take time, there is no way around this.