[
https://issues.apache.org/jira/browse/TOMEE-4571?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Martin A. updated TOMEE-4571:
-----------------------------
Description:
We recently migrated our PrimeFaces 13 application from TomEE9 to TomEE10 and
found a massive decrease in performance (Factor 2 or more).
*Measurement 1* : open dedicated page, then track its 3 XHR requests in browser
console
TomEE10 + JDK21 : 4900 ms - 6800 ms - 2100 ms
TomEE10 + JDK17 : 9200 ms - 8400 ms - 5700 ms
TomEE 9 + JDK17 : 2700 ms - 3000 ms - 0800 ms
TomEE 9 was > 2x faster, JDK might have a small impact, 21 is faster (as
expected)
As suggested by BalusC, i did a comparison with MyFaces 4.03 from TomEE Plus by
replacing it in TomEE Plume and ran the same testcases 3x to have reliable
results:
Measured times in ms, using p:lifecycle component (skipped phases < 10 ms).The
big differences are in RESTORE_VIEW phase = 10x faster and RENDER_RESPONSE
phase = 2-3x faster
TomEE 9 / mojarra 3.0.5
"search": RESTORE_VIEW 130 | INVOKE_APPLICATION 630 | RENDER_RESPONSE 850 |
"open dialog": RESTORE_VIEW 180 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE 400
|
TomEE 10 mojarra: 4.0.11
"search": RESTORE_VIEW *650* | INVOKE_APPLICATION 650 | RENDER_RESPONSE *1100* |
"open dialog": RESTORE_VIEW 750 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE 400
|
TomEE 10 myfaces: 4.0.3
"search" : RESTORE_VIEW 65 | INVOKE_APPLICATION 600 | RENDER_RESPONSE 650 |
"open dialog": RESTORE_VIEW 60 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE 150 |
MyFaces 4.0.3 is factor 2-10 faster than Mojarra 4.0.11, depending on action.
I also did a freaky try: run TomEE 10 with mojarra 3.0.5 (like TomEE 9) by
reverting namespace and related changed only, same tests as before (3x,same pc,
same db)
"Search": RESTORE_VIEW 700 | INVOKE_APPLICATION 650 | RENDER_RESPONSE 1200
"Open Dialog": RESTORE_VIEW 700 | INVOKE_APPLICATION 1400 | RENDER_RESPONSE 700
The idea that eclipselink / DB level is slower which turned out false. I
checked all server.xml, tomee.xml changes, as well as the required application
changes in dependencies, namespaces, faces-config.xml,
beans.xml persistence.xml and found no hint yet what could cause such decrease
in performance. I can confirm the idead that faces 3 and faces 4 are equal in
performance,
in fact "Restore view" phase is slower in 4. On the other hand, this now
means: There must be some other things in TomEE 10 that make it slower, e.g.
libraries, config changes, undocumented parameters, no-longer-in-use caches ...
So further Ideas are highly welcome.
was:
We recently migrated our PrimeFaces 13 application from TomEE9 to TomEE10 and
found a massive decrease in performance (Factor 2 or more).
Our Environment:
Primefaces 13.0.8, ojdbc11:21 , Oracle 19c database, gradle 8.7 build system.
PLUME edition of TomEE because we need eclipselink due to native queries and DB
package calls.
TomEE 9.1.3 PLUME as-is (Mojarra Faces 3, Eclipselink 3) on OpenJDK17
TomEE 10.1.2 PLUME as-is (Mojarra Faces 4, Eclipselink 4) on OpenJDK21
*Measurement 1* : open dedicated page, then track its 3 XHR requests in browser
console
TomEE10 + JDK21 : 4900 ms - 6800 ms - 2100 ms
TomEE10 + JDK17 : 9200 ms - 8400 ms - 5700 ms
TomEE 9 + JDK17 : 2700 ms - 3000 ms - 0800 ms
TomEE 9 was > 2x faster, JDK might have a small impact, 21 is faster (as
expected)
As suggested by BalusC, i did a comparison with MyFaces 4.03 from TomEE Plus by
replacing it in TomEE Plume and ran the same testcases 3x to have reliable
results:
Measured times in ms, using p:lifecycle component (skipped phases < 10 ms).The
big differences are in RESTORE_VIEW phase = 10x faster and RENDER_RESPONSE
phase = 2-3x faster
TomEE 9 / mojarra 3.0.5
"search": RESTORE_VIEW 130 | INVOKE_APPLICATION 630 | RENDER_RESPONSE 850 |
"open dialog": RESTORE_VIEW 180 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE 400
|
TomEE 10 mojarra: 4.0.11
"search": RESTORE_VIEW *650* | INVOKE_APPLICATION 650 | RENDER_RESPONSE *1100* |
"open dialog": RESTORE_VIEW 750 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE 400
|
TomEE 10 myfaces: 4.0.3
"search" : RESTORE_VIEW 65 | INVOKE_APPLICATION 600 | RENDER_RESPONSE 650 |
"open dialog": RESTORE_VIEW 60 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE 150 |
MyFaces 4.0.3 is factor 2-10 faster than Mojarra 4.0.11, depending on action.
I also did a freaky try: run TomEE 10 with mojarra 3.0.5 (like TomEE 9) by
reverting namespace and related changed only, same tests as before (3x,same pc,
same db)
"Search": RESTORE_VIEW 700 | INVOKE_APPLICATION 650 | RENDER_RESPONSE 1200
"Open Dialog": RESTORE_VIEW 700 | INVOKE_APPLICATION 1400 | RENDER_RESPONSE 700
The idea that eclipselink / DB level is slower which turned out false. I
checked all server.xml, tomee.xml changes, as well as the required application
changes in dependencies, namespaces, faces-config.xml,
beans.xml persistence.xml and found no hint yet what could cause such decrease
in performance. I can confirm the idead that faces 3 and faces 4 are equal in
performance,
in fact "Restore view" phase is slower in 4. On the other hand, this now
means: There must be some other things in TomEE 10 that make it slower, e.g.
libraries, config changes, undocumented parameters, no-longer-in-use caches ...
So further Ideas are highly welcome.
web.xml parameters overview
listener-class = com.sun.faces.config.ConfigureListener
servlet-class = jakarta.faces.webapp.FacesServlet
jakarta.faces.FACELETS_VIEW_MAPPINGS = .jsf;.xhtml
jakarta.faces.STATE_SAVING_METHOD = server
jakarta.faces.FACELETS_REFRESH_PERIOD = -1
jakarta.faces.PARTIAL_STATE_SAVING = true
jakarta.faces.FACELETS_BUFFER_SIZE = 1024000 (for fileuploads)
jakarta.faces.FACELETS_SKIP_COMMENTS = true
primefaces.CLIENT_SIDE_LOCALISATION = true
Environment:
Primefaces 13.0.8, ojdbc11:21 , Oracle 19c database, gradle 8.7 build system.
PLUME edition of TomEE because we need eclipselink due to native queries and DB
package calls.
TomEE 9.1.3 PLUME as-is (Mojarra Faces 3, Eclipselink 3) on OpenJDK17
TomEE 10.1.2 PLUME as-is (Mojarra Faces 4, Eclipselink 4) on OpenJDK21
web.xml parameters overview
listener-class = com.sun.faces.config.ConfigureListener
servlet-class = jakarta.faces.webapp.FacesServlet
jakarta.faces.FACELETS_VIEW_MAPPINGS = .jsf;.xhtml
jakarta.faces.STATE_SAVING_METHOD = server
jakarta.faces.FACELETS_REFRESH_PERIOD = -1
jakarta.faces.PARTIAL_STATE_SAVING = true
jakarta.faces.FACELETS_BUFFER_SIZE = 1024000 (for fileuploads)
jakarta.faces.FACELETS_SKIP_COMMENTS = true
primefaces.CLIENT_SIDE_LOCALISATION = true
> TomEE 10.1.x - Decrease in performance (Factor 2 or more).
> -----------------------------------------------------------
>
> Key: TOMEE-4571
> URL: https://issues.apache.org/jira/browse/TOMEE-4571
> Project: TomEE
> Issue Type: Bug
> Components: TomEE Core Server
> Affects Versions: 10.1.2
> Environment: Primefaces 13.0.8, ojdbc11:21 , Oracle 19c database,
> gradle 8.7 build system. PLUME edition of TomEE because we need eclipselink
> due to native queries and DB package calls.
> TomEE 9.1.3 PLUME as-is (Mojarra Faces 3, Eclipselink 3) on OpenJDK17
> TomEE 10.1.2 PLUME as-is (Mojarra Faces 4, Eclipselink 4) on OpenJDK21
> web.xml parameters overview
> listener-class = com.sun.faces.config.ConfigureListener
> servlet-class = jakarta.faces.webapp.FacesServlet
> jakarta.faces.FACELETS_VIEW_MAPPINGS = .jsf;.xhtml
> jakarta.faces.STATE_SAVING_METHOD = server
> jakarta.faces.FACELETS_REFRESH_PERIOD = -1
> jakarta.faces.PARTIAL_STATE_SAVING = true
> jakarta.faces.FACELETS_BUFFER_SIZE = 1024000 (for fileuploads)
> jakarta.faces.FACELETS_SKIP_COMMENTS = true
> primefaces.CLIENT_SIDE_LOCALISATION = true
> Reporter: Martin A.
> Priority: Major
>
> We recently migrated our PrimeFaces 13 application from TomEE9 to TomEE10 and
> found a massive decrease in performance (Factor 2 or more).
> *Measurement 1* : open dedicated page, then track its 3 XHR requests in
> browser console
> TomEE10 + JDK21 : 4900 ms - 6800 ms - 2100 ms
> TomEE10 + JDK17 : 9200 ms - 8400 ms - 5700 ms
> TomEE 9 + JDK17 : 2700 ms - 3000 ms - 0800 ms
> TomEE 9 was > 2x faster, JDK might have a small impact, 21 is faster (as
> expected)
> As suggested by BalusC, i did a comparison with MyFaces 4.03 from TomEE Plus
> by replacing it in TomEE Plume and ran the same testcases 3x to have reliable
> results:
> Measured times in ms, using p:lifecycle component (skipped phases < 10
> ms).The big differences are in RESTORE_VIEW phase = 10x faster and
> RENDER_RESPONSE phase = 2-3x faster
> TomEE 9 / mojarra 3.0.5
> "search": RESTORE_VIEW 130 | INVOKE_APPLICATION 630 | RENDER_RESPONSE 850 |
> "open dialog": RESTORE_VIEW 180 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE
> 400 |
> TomEE 10 mojarra: 4.0.11
> "search": RESTORE_VIEW *650* | INVOKE_APPLICATION 650 | RENDER_RESPONSE
> *1100* |
> "open dialog": RESTORE_VIEW 750 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE
> 400 |
> TomEE 10 myfaces: 4.0.3
> "search" : RESTORE_VIEW 65 | INVOKE_APPLICATION 600 | RENDER_RESPONSE 650 |
> "open dialog": RESTORE_VIEW 60 | INVOKE_APPLICATION 1200 | RENDER_RESPONSE
> 150 |
> MyFaces 4.0.3 is factor 2-10 faster than Mojarra 4.0.11, depending on action.
> I also did a freaky try: run TomEE 10 with mojarra 3.0.5 (like TomEE 9) by
> reverting namespace and related changed only, same tests as before (3x,same
> pc, same db)
> "Search": RESTORE_VIEW 700 | INVOKE_APPLICATION 650 | RENDER_RESPONSE 1200
> "Open Dialog": RESTORE_VIEW 700 | INVOKE_APPLICATION 1400 | RENDER_RESPONSE
> 700
> The idea that eclipselink / DB level is slower which turned out false. I
> checked all server.xml, tomee.xml changes, as well as the required
> application changes in dependencies, namespaces, faces-config.xml,
> beans.xml persistence.xml and found no hint yet what could cause such
> decrease in performance. I can confirm the idead that faces 3 and faces 4 are
> equal in performance,
> in fact "Restore view" phase is slower in 4. On the other hand, this now
> means: There must be some other things in TomEE 10 that make it slower, e.g.
> libraries, config changes, undocumented parameters, no-longer-in-use caches
> ... So further Ideas are highly welcome.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
