FWD » Core Development » Runtime Infrastructure

04/10/2016

06:22 PM Feature #2973: improve ContextLocal performance

I've restores the post-cleanup check.
Committed to the task branch 2973a revision 11006.
Automated regression t... Igor Skornyakov

04/09/2016

06:15 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> Constantin Asofiei wrote:
> > @tokenMap@ is guaranteed to not be changed by other threads.... Constantin Asofiei

03:28 AM Feature #2973: improve ContextLocal performance

I've performed majic regression testing of 2973a revision 11004.
# all CTRL-C tests passed
# from the main part the... Igor Skornyakov

04/08/2016

06:11 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> @tokenMap@ is guaranteed to not be changed by other threads. But is not guaranteed that... Igor Skornyakov

04:11 PM Feature #2973: improve ContextLocal performance

PS: if you take a look at @ContextLocal$Wrapper.cleanup@, you will see that after the var is cleaned, it is removed f... Constantin Asofiei

04:08 PM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> Sorry Constantin, I do not understand. If we are sure that @tokenMap@ is not changed during... Constantin Asofiei

03:51 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> > How this approach works with the situation when something was added to the @tokenMap@ ... Igor Skornyakov

03:35 PM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> How this approach works with the situation when something was added to the @tokenMap@ after... Constantin Asofiei

03:17 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> So, I think this should work:
> # this code is needed, as it is critical for @tokenMa... Igor Skornyakov

03:06 PM Feature #2973: improve ContextLocal performance

Igor, the code you commented from @cleanupWorker@ was responsible for two things:
# ensure that, after cleanup, @tok... Constantin Asofiei

12:52 PM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> After a close look at the @SecurityContext@ class I got an impression that if the @ tokenMa... Eric Faulhaber

12:07 PM Feature #2973: improve ContextLocal performance

After a close look at the @SecurityContext@ class I got an impression that if the @ tokenMap@ is a @ConcurrentMap@ we... Igor Skornyakov

11:57 AM Feature #2973: improve ContextLocal performance

Hynek Cihlar wrote:
> Igor Skornyakov wrote:
> > Hynek Cihlar wrote:
> > > Btw., have you considered making the lo... Igor Skornyakov

10:49 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> Hynek Cihlar wrote:
> > Btw., have you considered making the locking of reading operations... Hynek Cihlar

10:31 AM Feature #2973: improve ContextLocal performance

Hynek Cihlar wrote:
> Btw., have you considered making the locking of reading operations more optimistic? For exampl... Igor Skornyakov

10:22 AM Feature #2973: improve ContextLocal performance

Btw., have you considered making the locking of reading operations more optimistic? For example @StampLock@ allows th... Hynek Cihlar

09:08 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> The test run time is now 37 minutes 54 seconds.
> However the time spent in the @Security... Eric Faulhaber

08:54 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> However the time spent in the @SecurityContext.getToken()@ method decreased just to 55 seco... Constantin Asofiei

08:52 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
I've made the following changes:
# Replaced @ConcurrentHashMap@ with @HashMap@ as @tokenMa... Igor Skornyakov

07:13 AM Feature #2973: improve ContextLocal performance

I've used YourKit (sampling mode) to analyse server behaviour while running the customer test suite (**/*Test.java) (... Igor Skornyakov

06:09 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> This is a good idea. I will implement such test today. To make it more realistic - how may ... Constantin Asofiei

04:17 AM Feature #2973: improve ContextLocal performance

BTW if we are talking about a small map with is very infrequently updated but frequently read may be it makes sense t... Igor Skornyakov

04:02 AM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> Igor, following Eric's response, we can't go with the approach in note 37/38, as the num... Igor Skornyakov

03:45 AM Feature #2973: improve ContextLocal performance

Igor, following Eric's response, we can't go with the approach in note 37/38, as the number of @ContextLocal@ instanc... Constantin Asofiei

02:10 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> However if your tests show that @Map.get()@ is a bottleneck the reason can be that this map ... Eric Faulhaber

04/07/2016

06:02 PM Feature #2973: improve ContextLocal performance

Task branch 2973a was rebased from the trunk revision 10998. Pushed up to revision 11002. Igor Skornyakov

03:43 PM Feature #2973: improve ContextLocal performance

Greg Shah wrote:
>
> Hynek: can you report on @SyncCoordinatesAspect.contextData@ (see above discussion)?
I bel... Hynek Cihlar

03:23 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> Another headache is how to hide/ensure no one else can instantiate a @ContextLocal@ with... Igor Skornyakov

03:16 PM Feature #2973: improve ContextLocal performance

Greg Shah wrote:
> > This way, we can remove the map overhead completely. If the performance is still not improved, ... Constantin Asofiei

03:06 PM Feature #2973: improve ContextLocal performance

> This way, we can remove the map overhead completely. If the performance is still not improved, then the bottleneck ... Greg Shah

02:54 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> ... @tokenMap@ keys are always @ContextLocal@ instances, which are defined as static va... Constantin Asofiei

02:48 PM Feature #2973: improve ContextLocal performance

Greg, I think we should consider how many elements will @tokenMap@ hold, on maximum. @ContextLocal@ is instantiated ... Constantin Asofiei

02:40 PM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> Constantin Asofiei wrote:
> > The logic in @cleanupWorker@ does this:
> > * step 1: it so... Constantin Asofiei

02:31 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> The logic in @cleanupWorker@ does this:
> * step 1: it sorts or keys by their weight, w... Igor Skornyakov

02:26 PM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> OK, I will double check. Please not that it doesn't make sense to check the predicate twice... Constantin Asofiei

02:23 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> Followup for note 29: OTOH, this can be a fallback for the incorrect @cleanupWorker(bool... Igor Skornyakov

02:18 PM Feature #2973: improve ContextLocal performance

Eric Faulhaber wrote:
> In terms of performance, it was on the fastest end of the range of my recent timings for thi... Igor Skornyakov

02:07 PM Feature #2973: improve ContextLocal performance

Followup for note 29: OTOH, this can be a fallback for the incorrect @cleanupWorker(boolean reset, Map tm)@ filter co... Constantin Asofiei

02:02 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> Review for 2973a rev 10999:
> # the @tokenMapLock.readLock()@ and @tokenMapLock.writeLo... Igor Skornyakov

02:02 PM Feature #2973: improve ContextLocal performance

Eric Faulhaber wrote:
> I ran the full search test group. When I disconnected BPM at the end, while the P2J server w... Constantin Asofiei

01:54 PM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> I understand that NPE happens because FileSystemOps.WorkArea.fs is @null@. This can be of c... Eric Faulhaber

01:52 PM Feature #2973: improve ContextLocal performance

In terms of performance, it was on the fastest end of the range of my recent timings for this test group, but not dra... Eric Faulhaber

01:42 PM Feature #2973: improve ContextLocal performance

Eric,
I understand that NPE happens because FileSystemOps.WorkArea.fs is @null@. This can be of course a result of m... Igor Skornyakov

01:37 PM Feature #2973: improve ContextLocal performance

Review for 2973a rev 10999:
# the @tokenMapLock.readLock()@ and @tokenMapLock.writeLock()@ can be saved in @Security... Constantin Asofiei

01:31 PM Feature #2973: improve ContextLocal performance

Eric Faulhaber wrote:
> I ran the full search test group. When I disconnected BPM at the end, while the P2J server w... Igor Skornyakov

01:29 PM Feature #2973: improve ContextLocal performance

I ran the full search test group. When I disconnected BPM at the end, while the P2J server was still running, I got t... Eric Faulhaber

12:00 PM Feature #2973: improve ContextLocal performance

Greg Shah wrote:
> > However it seems that null is used in a single place: at the SecurityContextStack.setEditing me... Igor Skornyakov

11:52 AM Feature #2973: improve ContextLocal performance

> However it seems that null is used in a single place: at the SecurityContextStack.setEditing method. May be it is b... Greg Shah

11:32 AM Feature #2973: improve ContextLocal performance

Igor Skornyakov wrote:
> If we really need @null@ values we can wrap it into @Optional@. However it seems that @null... Eric Faulhaber

11:20 AM Feature #2973: improve ContextLocal performance

If we really need @null@ values we can wrap it into @Optional@. However it seems that @null@ is used in a single plac... Igor Skornyakov

10:48 AM Feature #2973: improve ContextLocal performance

Eric Faulhaber wrote:
> There is a regression when launching the P2J server:
> [...]
The ConcurrentHashMap doesn... Igor Skornyakov

10:46 AM Feature #2973: improve ContextLocal performance

There is a regression when launching the P2J server:... Eric Faulhaber

09:51 AM Feature #2973: improve ContextLocal performance

Eric Faulhaber wrote:
> Code review 2973a/10997:
>
> Yes, this is what I was looking for in the scope of note 5, ... Igor Skornyakov

09:49 AM Feature #2973: improve ContextLocal performance

I would like to see some testing that detects if this has a measurable performance impact (and what that might be).
... Greg Shah

09:19 AM Feature #2973: improve ContextLocal performance

Code review 2973a/10997:
Yes, this is what I was looking for in the scope of note 5, thanks.
One question: why ... Eric Faulhaber

08:52 AM Feature #2973: improve ContextLocal performance

Task branch 2973a was rebased from the trunk revision 10996. Committed as revision 10997. Igor Skornyakov

07:55 AM Feature #2973: improve ContextLocal performance

I've implemented optimization mentioned in the node 5 along with some refactoring.
Committed to the task branch 29... Igor Skornyakov

05:40 AM Feature #2973: improve ContextLocal performance

Created task branch 2973a from the trunk revision 10995. Igor Skornyakov

03/29/2016

03:37 PM Feature #3003: improve performance of BaseDataType.deepCopy()

> I'm not sure if this is the scenario that you're talking about in (6), I see that you're passing objects as referen... Greg Shah

03/25/2016

09:40 AM Feature #2973: improve ContextLocal performance

Thoughts:
1. @ContextLocal@ instances should always be stored in a static member. This allows for shared access t... Greg Shah

03/24/2016

01:34 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> For the @ContextLocal@ instance, is safe to use its @hashCode()@ implementation; the sam... Igor Skornyakov

01:24 PM Feature #2973: improve ContextLocal performance

Constantin Asofiei wrote:
> Eric Faulhaber wrote:
> > For this to work, there has to be a secure and fast mechanism... Eric Faulhaber

12:58 PM Feature #2973: improve ContextLocal performance

Eric Faulhaber wrote:
> For this to work, there has to be a secure and fast mechanism to get the current security co... Constantin Asofiei

12:37 PM Feature #2973: improve ContextLocal performance

The more complicated approach would be something to consider if the synchronization improvements described in the pre... Eric Faulhaber

12:12 PM Feature #2973: improve ContextLocal performance

Greg and I have been discussing several approaches to this. The simpler of these, which may buy us back sufficient pe... Eric Faulhaber

09:05 AM Feature #2973: improve ContextLocal performance

I see. Thank you for the detailed explanations. Igor Skornyakov

09:02 AM Feature #2973: improve ContextLocal performance

To be clear, we are specifically talking about optimizing the @MODE_SERVER@ mode of @ContextLocal@, which uses the @S... Greg Shah

01:00 AM Feature #2973 (WIP): improve ContextLocal performance

Igor, please familiarize yourself with the the current @ContextLocal@ implementation and think about how to best impl... Eric Faulhaber