Instrumented profiling for multimodule Jigsaw startup
Aleksey Shipilev
aleksey.shipilev at oracle.com
Thu Mar 15 05:52:52 PDT 2012
On 03/15/2012 04:36 PM, Alan Bateman wrote:
> On 14/03/2012 22:16, Aleksey Shipilev wrote:
>> :
>> The thing I missed from explanation there: if you churn those 500ms
>> local class map loading out of 600ms of total startup time,
>> modularized Jigsaw startup is on-par with "legacy" JDK8 startup for
>> HelloWorld tests. Do you want me to see what could be addressed in
>> configuration, or is it someone else's part already?
>>
>> -Aleksey.
> A while back (perhaps >1 year ago) then Mark had a prototype of a fast
> configuration that used a mini database. That work needs to be re-based
> and pushed to jigsaw/jigsaw. I see your other mail with a patch to
> improve the existing format/code, much appreciated and we can look at
> that for the short term but eventually this will be replaced.
Is this the part of that mini-database stuff?
http://cr.openjdk.java.net/~mr/jigsaw/notes/module-file-format/
Are we sure mini database (whatever that means) would be faster than
carefully-tuned in-memory Java collections and efficient
marshallers/demarshallers, tied to hand-tuned on-disk form?
I have a reason to believe it would not make any difference at this
point. We are talking about ~30ms time spent in configuration loader
after my "reverse map" patch, in absence of pre-compiled classes,
running in interpreted mode, experiencing huge I/O, etc.
-Aleksey.
More information about the jigsaw-dev
mailing list