How do you install the large-scale Alfresco CIFS server?

Question

Alfresco provides a CIFS connector, so it can act like a regular file server on your intranet.

Compared to a "normal" (windows / samba) file server, certain operations can really damage the system, for example. listing a folder with several thousand files using Windows Explorer. Not quite sure, but I believe that access control is the main reason for this case. In any case, now suppose you have a large file system hierarchy, and many users use CIFS, emphasizing the system, effectively "knocking it down."

What is the proposed scaling / performance improvement approach?

Answer 1

In my experience, Windows Explorer is part of the CIFS performance issue. I don’t have exact numbers, but I remember that I worked on an instance with approximately 500 GB of data, mainly consisting of small images and several texts in a not very balanced folder tree, for which listing a folder with a thousand children was taken in Explorer for about a minute to display . The same operation took about 3 seconds in the Chrome browser.

We did not have time to carefully study the problem, but we saw an impressive amount of traffic generated by Explorer, due to the preliminary selection of information about the subfolders of the open folder.

Answer 2

I’m reviewing the problem a bit, and I think that the best answer I can give right now is to configure caching.

I used 5 thousand free space, default cache values ​​and control execution of " ls -alrt " on CIFS mount under alfresco 4.0.d.

The first execution took about two minutes, bombarding (lightning fast) the mysql database with queries about 200 thousand.

The second execution took "only" about 40 seconds, but the number of requests did not change significantly.

By increasing the cache file in the CIFS file, I got the second time up to 30 seconds, but I still see how the 160 KB request is launched. I am sure that this share of lions is related to ACL permissions / lists, and this should greatly improve the situation.

PS: Windows Explorer definitely behaves a little unexpectedly, but I can’t confirm that it is significant to the user.

PPS: https://issues.alfresco.com/jira/browse/ALFCOM-2951

PPPS: I will consider this further when I find the time - it should be this year .;)

Update. Massive number of requests without permission .

Answer 3

Checking permissions is definitely part of the problem. I cannot refer to anything specific, but by browsing the open air forums and the network over the past few years, I have learned that permissions can hurt performance.

I read (and experienced) in several scenarios that spaces with alfresco with lots of children (1000+) can be painfully slow. One part that you yourself noticed: it takes some time to complete 100-200k requests. But plug something into the street to see what it does, and you will see that a huge amount of time goes for serialization / deserialization (egwebscripts for share), as well as node crawl (hence thousands of requests and average values ​​of 400-500 qps when no one is logged in). Thus, you are configured correctly with cache optimization.

Do you have special equipment for your installation? I had big performance problems, but I moved the MySQL server to a separate block (server-level hardware - 4 cores, 8 GB of RAM, SSD for myqsl server and SAS for tomcat server, etc.), and I got a lot . So, keep asking for new equipment :)

I think you are on the right track here.