Hello Alexey, Thanks for your answer! I was wondering about reproducers to submit against ICU and thought that the ones from JIRAs depend from our implementation as well.
> There are some hidden performance problems which were exposed during ICU upgrade, some of which we can handle after reaching the milestone. This includes at least proper caching of useDaylightTime property and using simpler ICU's useDaylightTime(). Aha, I see. So if I got this correctly this problem could be fixed on our side by means of caching. Returning to my initial question about the plan, it seems that the second item should look like "Invest time into proper ICU adoption" instead of "File all the found issues to ICU database". Thanks. On Nov 18, 2007 7:30 PM, Aleksey Shipilev <[EMAIL PROTECTED]> wrote: > Hi, Tony, Alexei! > > > Following commits are reverted at r596044. > Thanks, Tony! > > > I have looked through JIRAs and did not find a reproducer. Do we have one? > You might refer to these two JIRAs dedicated for performance > degradations: [1] and [2]. There you will find the microtests on > problematic areas of changed code and their results. > > > > useDaylightTime > > How this method might cause server benchmark degradation? Who calls > > this method frequently? May be we should change the caller either it > > is a part of the spec or our implementation? > It is used for internal computation in j.u.Date and > j.u.GregorianCalendar. As soon as you are running buisiness workload, > you will have to handle the Date too, hence the workload performance > degradations caused by j.u.Date degradations. > > There are some hidden performance problems which were exposed during > ICU upgrade, some of which we can handle after reaching the milestone. > This includes at least proper caching of useDaylightTime property and > using simpler ICU's useDaylightTime(). > > Thanks, > Aleksey, > ESSD, Intel > > [1] http://issues.apache.org/jira/browse/HARMONY-5122 > [2] http://issues.apache.org/jira/browse/HARMONY-5129 > -- With best regards, Alexei, ESSD, Intel
