[xwiki-devs] Myxwiki.org farm performance, when are we full?
Hi devs, We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it. When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time? Thanks -Vincent
Hi Vincent, Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time. We don't have tools that detect usages per wiki AFAIK. Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines. Jerome. On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol <[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
On Jan 28, 2011, at 12:02 PM, Jerome Velociter wrote:
Hi Vincent,
Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time.
That would be great. What's interesting is HTTP Response time over time (ie for example every month). That should allow us to see if perf is degrading or not. Of course the causes could be multiple: - too many wikis - perf issues with an XE/XEM release - a wiki taking too much resources on a new page
We don't have tools that detect usages per wiki AFAIK.
Having the stats in a per wiki basis would be great. Isn't that possible since the domains are not the same it should be easy to show stats for each, no?
Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines.
Ah cool didn't know that. Thanks -Vincent
Jerome.
On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol <[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
On 01/28/2011 12:02 PM, Jerome Velociter wrote:
Hi Vincent,
Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time.
We don't have tools that detect usages per wiki AFAIK.
We can extract some (edit) statistics from the activity events. Here's a short summary: http://www.myxwiki.org/xwiki/bin/Admin/FarmActivityReport
Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines.
Maybe the current machine could be reinstalled from scratch, since the mysql socket problem isn't solved yet. Reminder, the problem I'm talking about is that in most myxwiki restart logs, mysql processlist shows open connection that are in a NULL state (they're done executing the query and sending the response), while java lists several threads blocked on reading data from mysql, data that won't ever come since mysql considers it did all its work.
Jerome.
On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol<[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
-- Sergiu Dumitriu http://purl.org/net/sergiu/
On Jan 28, 2011, at 3:44 PM, Sergiu Dumitriu wrote:
On 01/28/2011 12:02 PM, Jerome Velociter wrote:
Hi Vincent,
Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time.
We don't have tools that detect usages per wiki AFAIK.
We can extract some (edit) statistics from the activity events. Here's a short summary:
Nice, probably worth a code snippet on extensions.xwiki.org ;) When it says "Events" on that page does it include all events (including view events) or only some events (like only edit events)? However the issue is that we can't really translate the number of events with the wiki farm's load, can we? Thanks -Vincent
Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines.
Maybe the current machine could be reinstalled from scratch, since the mysql socket problem isn't solved yet.
Reminder, the problem I'm talking about is that in most myxwiki restart logs, mysql processlist shows open connection that are in a NULL state (they're done executing the query and sending the response), while java lists several threads blocked on reading data from mysql, data that won't ever come since mysql considers it did all its work.
Jerome.
On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol<[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
On 01/28/2011 04:56 PM, Vincent Massol wrote:
On Jan 28, 2011, at 3:44 PM, Sergiu Dumitriu wrote:
On 01/28/2011 12:02 PM, Jerome Velociter wrote:
Hi Vincent,
Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time.
We don't have tools that detect usages per wiki AFAIK.
We can extract some (edit) statistics from the activity events. Here's a short summary:
Nice, probably worth a code snippet on extensions.xwiki.org ;)
Yes, I was going to do that after I add some links in it.
When it says "Events" on that page does it include all events (including view events) or only some events (like only edit events)?
Only events stored by the activity stream. It doesn't include view events, it used to store only create/update/delete document events, but since 2.6 it also stores comment/attachment/annotation events.
However the issue is that we can't really translate the number of events with the wiki farm's load, can we?
No, that's not a very good indicator of the wiki maintainance. Some wikis could have plenty of view, although little edits. But it's a starting point in determining which wikis should be parked. Still, a wiki with no activity wouldn't be such a problem for the farm, since it doesn't consume a lot of resources. On the contrary, most active wikis should be moved to their own server.
Thanks -Vincent
Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines.
Maybe the current machine could be reinstalled from scratch, since the mysql socket problem isn't solved yet.
Reminder, the problem I'm talking about is that in most myxwiki restart logs, mysql processlist shows open connection that are in a NULL state (they're done executing the query and sending the response), while java lists several threads blocked on reading data from mysql, data that won't ever come since mysql considers it did all its work.
Jerome.
On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol<[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
-- Sergiu Dumitriu http://purl.org/net/sergiu/
Here are a first set of graphs, exported statically from cacti : http://myxwiki.org/xwiki/bin/view/Admin/Graphs It includes, hourly, daily, weekly, monthly and yearly averages for : HTTP response time MySQL connections MySQL queries TCP Response time Tomcat heap (free and used) Tomcat connection rate (requests and errors) Threads (existing and busy) Tomcat throughput (inbound and outbound) This one is static, but we'll try to have dynamic graphs directly in wiki pages in a near future. We'll keep you posted. Jerome, on behalf of XWiki SAS platform team On Fri, Jan 28, 2011 at 12:02 PM, Jerome Velociter <[email protected]> wrote:
Hi Vincent,
Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time.
We don't have tools that detect usages per wiki AFAIK.
Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines.
Jerome.
On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol <[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
Hi Jerome, Thanks for this! I tried to analyze quickly but I couldn't draw any conclusion. Some data look dubious. For example the HTTP response time: they say that no HTTP request took longer than 40-50ms. I don't believe this or I don"t understand what it means. Loading a page takes more than 1 seconds overall, spread over several HTTP request but the main request itself takes more than 1 second. Also since myxwiki.org restarts automatically when it cannot access it through an HTTP request for more than 30 seconds, we should see HTTP request taking 30 seconds! Any idea? Thanks -Vincent On Feb 1, 2011, at 11:54 AM, Jerome Velociter wrote:
Here are a first set of graphs, exported statically from cacti :
http://myxwiki.org/xwiki/bin/view/Admin/Graphs
It includes, hourly, daily, weekly, monthly and yearly averages for :
HTTP response time MySQL connections MySQL queries TCP Response time Tomcat heap (free and used) Tomcat connection rate (requests and errors) Threads (existing and busy) Tomcat throughput (inbound and outbound)
This one is static, but we'll try to have dynamic graphs directly in wiki pages in a near future. We'll keep you posted.
Jerome, on behalf of XWiki SAS platform team
On Fri, Jan 28, 2011 at 12:02 PM, Jerome Velociter <[email protected]> wrote:
Hi Vincent,
Yes, we can publish some Cacti graphs, for the overall CPU and memory usage of the machine and HTTP response time.
We don't have tools that detect usages per wiki AFAIK.
Note that we plan to add a new machine for myxwiki soon, and configure a load balancer in front of the 2 machines.
Jerome.
On Fri, Jan 28, 2011 at 10:36 AM, Vincent Massol <[email protected]> wrote:
Hi devs,
We're accepting new wikis on myxwiki.org as requests come in but we already have 110 wikis created on it.
When do we know we shouldn't accept new wiki creation on that JVM because it's degrading performances too much? Do we have tool to monitor wiki activity and publish on a page which wikis are using the most resources? Do we have tools to display response time stats and see how they evolve over time?
Thanks -Vincent
participants (3)
-
Jerome Velociter -
Sergiu Dumitriu -
Vincent Massol