Conversation starter on Confluence issues/questions
unixg...@gmail.com
to discuss. More than just the ones in existing JIRA issues. I thought
I'd throw some things out there. If people find them interesting, we
can spin off independent threads.
1. How do you measure performance on your production system in a non-
invasive way? For us, confluence access logs don't capture the end-
user results. The JMeter scripts can't be used on production, and
don't give good results on our staging systems. A developer has tried
Hyperic with little luck, and our Unix server monitoring tools show
that Confluence is only using a fraction of the resources available.
Low load and such.
2. Do you allow users to create their own view restrictions on
content, and how do you get them to do it correctly? I'm thinking of a
scenario where someone asks for a group called MyCompany, creates a
space called MyCompany but doesn't restrict it right, and then starts
posting company proprietary stuff without realizing the entire
Internet can read it, and any registered user can change it.
3. Do people outside your network have much slower performance, and
have you done anything to improve this?
4. We've redesigned our main site so that the dashboard is not the
initial page. However, we still have a lot of slowness whenever first
logging in. Have you experienced this, and have you done anything
about this? Our other site still uses the dashboard with 400 spaces,
so I'd expect it to be slow...
5. How have you handled requests from groups that wanted a
particularly specialized wiki? Or came in with a mock-up that
resembled a website?
6. How do you keep plugins that one group wants from impacting the
performance of the whole site?
7. Have you hired external help for any performance improvement, or
have you worked with Atlassian tech support, or done it on your own?
8. For sites accessible via anonymous access to the whole Internet, do
you restrict crawling the site and/or putting it up on search sites
like Google? And does it cause a performance hit while being indexed?
9. Do people actually pay attention to tutorials you create?
10. Have you found that page restrictions or space restrictions cause
performance problems?
11. What particular plugins have you had problems with? Particularly
third party ones.
12. Does your company have multiple installations of confluence, and
what has driven you to do so?
Mark Nye (UIUC)
On Apr 22, 3:38 pm, "unixgirl...@gmail.com" <unixgirl...@gmail.com>
wrote:eads.
> 1. How do you measure performance on your production system in a non-
> invasive way?
load time as a 20 minutes moving average. We also graph slowest page
load and fastest page load times per 5 minutes. This is almost useless
as an absolute performance indicator, but does show us relative
performance, and gives us a clue when we're encountering some issues.
We also keep an eye out for full GC events, and a service is notified
if we see more than one in a 5 minutes period.
>
> 2. Do you allow users to create their own view restrictions on
> content, and how do you get them to do it correctly?
For the most part we haven't run into problems, though the page move
restrictions bug has bitten us a couple of times.
https://jira.atlassian.com/browse/CONF-4111
> 3. Do people outside your network have much slower performance, and
> have you done anything to improve this?
specifically related to off-campus Confluence performance.
> 4. We've redesigned our main site so that the dashboard is not the
> initial page. However, we still have a lot of slowness whenever first
> logging in. Have you experienced this, and have you done anything
> about this? Our other site still uses the dashboard with 400 spaces,
> so I'd expect it to be slow...
get dog slow. What kind of replacement page did you design? How do you
generate space lists for your users?
> 5. How have you handled requests from groups that wanted a
> particularly specialized wiki? Or came in with a mock-up that
> resembled a website?
space layouts. It's my opinion that a CMS is better suited to fulfill
this need, but I can understand why some people would be interested in
using Confluence.
> 6. How do you keep plugins that one group wants from impacting the
> performance of the whole site?
environment. If the plugin is found to have a security issue or
present performance problems, we won't install it to production. When
we do go to production with a new plugin, we inform the users that it
may need to be removed, if it causes an unexpected performance hit.
> 7. Have you hired external help for any performance improvement, or
> have you worked with Atlassian tech support, or done it on your own?
Atlassian, and with input from other academic users. We've not engaged
consultants.
> 8. For sites accessible via anonymous access to the whole Internet, do
> you restrict crawling the site and/or putting it up on search sites
> like Google? And does it cause a performance hit while being indexed?
indexing, but this decision may soon need to be reevaluated.
> 9. Do people actually pay attention to tutorials you create?
> 10. Have you found that page restrictions or space restrictions cause
> performance problems?
> 11. What particular plugins have you had problems with? Particularly
> third party ones.
we're looking at removing it from our system. I believe that the
calendar plugin may also contribute to some service slowdowns, but I
haven't substantiated that. The usage tracking plugin is known to
cause problems in large installations, but we've not experienced
issues with it.
> 12. Does your company have multiple installations of confluence, and
> what has driven you to do so?
Confluence install can't be scaled to meet the demands of the campus.
best,
Mark
unixg...@gmail.com
that interest you or come up with your own conversation starters.
On Apr 22, 5:07 pm, "Mark Nye (UIUC)" <mark...@uiuc.edu> wrote:
> On Apr 22, 3:38 pm, "unixgirl...@gmail.com" <unixgirl...@gmail.com>
> wrote:eads.
>
>
>
> > 1. How do you measure performance on your production system in a non-
> > invasive way?
>
> We've enabled page request logging in confluence, and we graph page
> load time as a 20 minutes moving average. We also graph slowest page
> load and fastest page load times per 5 minutes. This is almost useless
> as an absolute performance indicator, but does show us relative
> performance, and gives us a clue when we're encountering some issues.
>
> We also keep an eye out for full GC events, and a service is notified
> if we see more than one in a 5 minutes period.
>
>
http://confluence.atlassian.com/display/DOC/How+to+audit+Confluence+-+enabling+user+access+logging
On 2.8 it doesn't give page edit/create times. That is one of our
concerns. We've considered hacking the filters but have other
priorities.
Over long, long periods of time (weeks/months) the averages can
provide trends. Our response times minute to minute seem to vary a
lot, so I don't know if a 20 minute moving average would hlep.
> > 3. Do people outside your network have much slower performance, and
> > have you done anything to improve this?
>
> No. To the best of my knowledge, we've never received a complaint
> specifically related to off-campus Confluence performance.
>
both here and across town. Typically several seconds slower. And
inevitably whenever there is an important demo someplace our site is
slow.
The user's "slowness" experience is not always correlated to the
access logs.
> > 4. We've redesigned our main site so that the dashboard is not the
> > initial page. However, we still have a lot of slowness whenever first
> > logging in. Have you experienced this, and have you done anything
> > about this? Our other site still uses the dashboard with 400 spaces,
> > so I'd expect it to be slow...
>
> We're still using the default dashboard layout, but it's starting to
> get dog slow. What kind of replacement page did you design? How do you
> generate space lists for your users?
>
set up, and people enter it.
Create an index.html on the main webserver URL that redirects people
to another page.
Set up the Confluence welcome message to contain redirect code to
whatever you want.
Use http://confluence.atlassian.com/display/DOC/Redirect+users+to+a+page+on+login
to do something.
Make sure everyone goes to the favorites page. The first time they log
in, ask what their role is: Undergrad/Grad/Faculty/Researcher/Staff.
On our big install, favorite spaces are really the way to go. Even
that can be slow.
This is a random suggestion I'm throwing out. I'm probably not using
the right terms or doing things that are technically possible.
I'd do something like creating a custom page that inserts the user's
favorite spaces. Or have the custom page show "new spaces" at the
bottom.
Then you associate each space with one or more tags. To search the
spaces, the person could select tags. Maybe have most of the tags be
pre-selected. And the space creator could create one. So the space for
the Society of Women Engineers could be tagged, "engineering school",
"student organization" "women" "undergraduate" "graduate".
Let the user check the tags they are interested in.
> > 5. How have you handled requests from groups that wanted a
> > particularly specialized wiki? Or came in with a mock-up that
> > resembled a website?
>
> Yes, and most of our folks are using Theme Builder to customize their
> space layouts. It's my opinion that a CMS is better suited to fulfill
> this need, but I can understand why some people would be interested in
> using Confluence.
>
particular group. Look and feel, pre-populate some content, etc. So
they can get on with their jobs and just add to it.
13. How large of a support staff do you have? Dedicated, part time,
etc.
14. What kind of page response time do you typically see? Not just in
the confluence access logs, but if you had a user with a stopwatch, or
use YSlow on Firefox? What is a realistic expectation? Do you get
users saying "You aren't as fast as Wikipedia and why aren't you on
MediaWiki instead of Confluence?"
Thanks,
Alice
Igor Minar
On Apr 22, 2009, at 3:17 PM, unixg...@gmail.com wrote:
>
> People don't have to answer all the questions - just respond to things
> that interest you or come up with your own conversation starters.
>
> On Apr 22, 5:07 pm, "Mark Nye (UIUC)" <mark...@uiuc.edu> wrote:
>> On Apr 22, 3:38 pm, "unixgirl...@gmail.com" <unixgirl...@gmail.com>
>> wrote:eads.
>>
>>
>>
>>> 1. How do you measure performance on your production system in a
>>> non-
>>> invasive way?
>>
>> We've enabled page request logging in confluence, and we graph page
>> load time as a 20 minutes moving average. We also graph slowest page
>> load and fastest page load times per 5 minutes. This is almost
>> useless
>> as an absolute performance indicator, but does show us relative
>> performance, and gives us a clue when we're encountering some issues.
>>
>> We also keep an eye out for full GC events, and a service is notified
>> if we see more than one in a 5 minutes period.
Depending on your webserver you might be able to turn request duration
logging in your access log.
I also find it useful to get an idea of the current performance by
looking at the request execution JMX stats. Confluence is instrumented
via JMX, you knew that, right?
I still would like to see more stats exposed via JXM, e.g.
http://jira.atlassian.com/browse/CONF-14912
http://jira.atlassian.com/browse/CONF-14913
>>
>>
> Are you talking about this?
> http://confluence.atlassian.com/display/DOC/How+to+audit+Confluence+-+enabling+user+access+logging
>
> On 2.8 it doesn't give page edit/create times. That is one of our
> concerns. We've considered hacking the filters but have other
> priorities.
>
> Over long, long periods of time (weeks/months) the averages can
> provide trends. Our response times minute to minute seem to vary a
> lot, so I don't know if a 20 minute moving average would hlep.
>
>
>>> 3. Do people outside your network have much slower performance, and
>>> have you done anything to improve this?
>>
>> No. To the best of my knowledge, we've never received a complaint
>> specifically related to off-campus Confluence performance.
>>
>
> Darn. A group across town has done regular testing on the same pages
> both here and across town. Typically several seconds slower. And
> inevitably whenever there is an important demo someplace our site is
> slow.
>
> The user's "slowness" experience is not always correlated to the
> access logs.
Sounds like a network problem to me. Try using ping/traceroute to
discount this possibility.
That's hard to answer. I'm the lead eng spends majority of my time on
the wikis.sun.com project. There are other engineers who pitch in from
time to time and we have a shared DBA and operations teams that are in
charge of databases, os, hw and network.
>
>
> 14. What kind of page response time do you typically see? Not just in
> the confluence access logs, but if you had a user with a stopwatch, or
> use YSlow on Firefox? What is a realistic expectation? Do you get
> users saying "You aren't as fast as Wikipedia and why aren't you on
> MediaWiki instead of Confluence?"
The initial load is terrible because confluence has close to one
million asset files which get loaded (ok, I'm kidding, but there are
way too many individual css/js files :-) ). AFAIK in Conf3.0 this was
finally fixed by concatenating and minifying these files.
Consecutive page loads are much faster because most of the asset files
are cached in the browser. Still there are some images that are not
sent with the proper caching headers and are fetched with each request
(editor, emoticons, some other icons).
After this the individual page load time depends on the number and
kind of macros used on a page, size of the page, complexity of ACL,
state of the coherence cache, https and many other factors.
Having said this, in general we see total end user page load times
somewhere between 1-7 seconds.
The page load time could be much better if the renderer was more
intelligent. Did you know that every confluence page is translated
from wiki markup and rendered almost from scratch during each page
load? There is so much opportunity for caching here, especially if
macros were required to define if their output could be cached.
/i
Igor Minar
On Apr 22, 2009, at 1:38 PM, unixg...@gmail.com wrote:
> 2. Do you allow users to create their own view restrictions on
> content, and how do you get them to do it correctly? I'm thinking of a
> scenario where someone asks for a group called MyCompany, creates a
> space called MyCompany but doesn't restrict it right, and then starts
> posting company proprietary stuff without realizing the entire
> Internet can read it, and any registered user can change it.
We patched confluence to change the ACL defaults when a space is
created.
This could be also achieved via a plugin that listens to space
creation events.
We also display a warning banner on the edit page if a page is public.
>
> 3. Do people outside your network have much slower performance, and
> have you done anything to improve this?
>
> 4. We've redesigned our main site so that the dashboard is not the
> initial page. However, we still have a lot of slowness whenever first
> logging in. Have you experienced this, and have you done anything
> about this? Our other site still uses the dashboard with 400 spaces,
> so I'd expect it to be slow...
>
> 5. How have you handled requests from groups that wanted a
> particularly specialized wiki? Or came in with a mock-up that
> resembled a website?
no customizations are allowed at our site.
>
> 6. How do you keep plugins that one group wants from impacting the
> performance of the whole site?
we are *very* conservative about plugin installation, especially when
it comes to unsupported plugins.
>
> 7. Have you hired external help for any performance improvement, or
> have you worked with Atlassian tech support, or done it on your own?
on our own
>
> 8. For sites accessible via anonymous access to the whole Internet, do
> you restrict crawling the site and/or putting it up on search sites
> like Google? And does it cause a performance hit while being indexed?
http://wikis.sun.com/robots.txt
>
> 9. Do people actually pay attention to tutorials you create?
I hope they do! :)
>
> 10. Have you found that page restrictions or space restrictions cause
> performance problems?
No, but I heard about this from people running other confluence
instances.
>
> 11. What particular plugins have you had problems with? Particularly
> third party ones.
Most of the community plugins are not suitable for instances like ours
either due to security holes or scalability issues.
>
> 12. Does your company have multiple installations of confluence, and
> what has driven you to do so?
This was proposed to us a solution to our scalability issues, but we
turned it down due to the fact that it is not solving the real issues,
adds maintenance overhead and causes user confusion.
/i