Slow performance

[fschnell]

We have Alfresco 2.2E installed and uploaded 120,000 documents residing in a hierarchy of a handfull (maybe 100) spaces. Access via CIFS or the webinterface is fast on the first level showing about 15 spaces. From there opening a space with 200 documents takes about 10 seconds, with 500 documents >30 seconds and every space with more than that is inaccessible. There are 150 LDAP users defined and the backend database on a separate server is Oracle 9. I got the recommendation not to store more than 200 documents in a space/folder as performance after that would massively decrease.

As I do not know where to start analysing and what values would be normal or common, can others please post access times for folders with various amounts of documents, e.g. 100 files, 200 files, 500 files and a 1000 files; what database is used and whether it is on a separate server.

Thanks for your help
Frank

[alahser]

Hi Frank,

I'm new to alfresco too. I'm not sure what database you are using, but, I would bet that your database is sorting those longer lists of files in a temporary table that is written to disk. So, you might want to try to increase the memory that your database uses for temporary tables. In MYSQL, you might want to increase max_heap_table_size and tmp_table_size. I just kept doubling these values until performance was acceptable.

You may also have fragmented indexes or other problems if you have not done any DB maintenance since loading the your 120,000 documents. Moreover, I have found that tuning the database and giving the Alfresco JVM additional memory the best way to deal with performance issues.

Best Regards,
Andrew

[itbeb]

We have the same problem. What browser are you using? Look at the difference in response time between IE and Firefox - in Firefox everything is displayed instantaneouly as opposed to the time differences in IE. Interesting huh?  :roll:

Regards

[fschnell]

Look at the difference in response time between IE and Firefox - in Firefox everything is displayed instantaneouly

Well, it is difficult to believe that this is a browser problem. The difference between IE and FF renderimg is within a few milliseconds, maybe one second. I am talking about a delay of 20-30 seconds (that's 20,000 ms). And the problem clearly shows on the CIFS interface too, if not even worse.

(side note: Opera beats both of them in terms of rendering speed. Just try it once!)

[skorde]

Hi
We are also facing the same problem with our production  environment(Alfresco Community 2.1+Oracle 9i)
We have around 6,67,743 folder and document already uploaded.
We are not able to understand why Alfresco taking so much time to upload a new file through CIFS ?
earlier it's faster now a day's we are facing performance issues?
We want to know what is the root cause?

Thanks in Advanced

[fschnell]

One reason, I was told, is, that Alfresco has to verify the access rights for each and every file to be displayed. A rather time consuming exercise.

[dgenard]

Hi Frank,

we've a very similar performance problems when opening large spaces.
Opening a space with 970 subspaces takes about 35 secs in Windows explorer (CIFS), and 80 secs in Alfresco Web UI.

These figures are measured on a single CPU Alfresco 2.1.1 / MySQL / Windows 2003 32-bits.
Up to now, we tried rebuilding indexes, but it has no significant impact.

I'll try tuning other parameters :
- JVM parameters
- Upgrade to latest 2.1 SP5 (to fix the fragmented indexes bug)
- Upgrade to 2.2 SP1 (they changed the way to store and check permissions)
- MySQL max_heap_table_size and tmp_table_size
and keep you informed of results.

Do someone have other ideas or significant results to improve these performances ?

Regards, Denis

Post-editing - Results of our tests :
None of above parameters have a significant impact on performace (JVM parameters, 2.2SP5, 2.2SP1).
MySQL max_heap_table_size and tmp_table_size not tested, but CPU powser is consumed almost exclusively by tomcat during such time intensive operations.

However, during testing, we noticed that caches can dramatically improve performances (successive opening of large space : 80 sec first time, then about 6 sec).
My conclusion is that caching is definitely an important factor to consider.
I'll further study how to better tune caching parameters (DB level, level 1, level 2).
Any experience feedback on this would be welcome.

Denis

[tarun_kapoor]

Our production environment consists of Alfresco community 2.1 and Oracle 9i. We have around 7 lacs folders and documents uploaded in Alfresco. This instance is taking alot of time for precisely all the activities like opening, saving, uploading a document, etc. Response time has got pretty bad in the recent times. In some cases we have to wait for more than a minute. We are using WebDAV, Web and CIFS interfaces for accessing Alfresco. We have not seen any performance degradation earlier when the repository size was small. JVM size is 1024 MB and system is deployed on a blade server with dual core CPU having 4 GB RAM and database is on a separate same configuration blade server. We have also found through Alfresco forum posts that people on Enterprise 2.2 have also faced this issue.

Can anyone suggest a possible remedy for this?

[munwar]

We have implemented Alfresco for large storage (millions of documents; with around 10 TB of data and meta-data) in Production environment.  However we had a clustered deployment to manage the load and to give us good performance.

There might be various reasons for slow performance which are both Alfresco related and non-Alfresco related. You need to locate the actual problem.

Non-Alfresco related:
- Network issues
- Database connectivity issues
- File System Access issues
- Slower LDAP or Active Directory (if used)
- Server (CPU) Power etc…
- Number of people using Alfresco at a specific point of time (large number of concurrent users)

Alfresco Related:

- File operations (add/edit/delete) in a space which has LARGE number of files takes more time.

- Having versionable aspect on new nodes has a huge negative impact on performance. Some noticed more than 50% improvement in performance by not applying versioning.

- File operations in a space which has business rules (run in foreground not in background) takes up good amount of time.  For example the rules such as transforming content from Word to PDF, executing custom Javascript etc.

- Unbalanced deployment of Alfresco (need to check the topology of your production deployment)

- Unsupported Stack of Alfresco (Please check Alfresco website for the stack supported)

- Community version or Older version which has known performance related bugs (Use the latest Enterprise tested version)

- Too much of logging the details (see if you have DEBUG option enabled in Production environment)

- Enabled Audit Trail (By default this is OFF), which does lots of extra activities for each operation

- the list goes on … depending upon the way you implemented Alfresco

Review your production deployment for such situations.

Alfresco is deployed for large production environments (for 80 million page views / month, for 10 TB of data etc..), hence it is a proven stable platform for enterprise roll-outs.  May be consider getting a supported (Enterprise) version which saves lots of your time.

[beren]

That's a great response. I'm also looking into performance issues with our implementation. Is there some kind of guide to how to tune alfresco? In my experience most J2EE projects will have a whitepaper or a manual on what settings are important to look at, examples , caching tools and configurations, and stuff like that. Is there anything like this for alfresco?

beren