Here are some notes about the deployment of the Alfresco server at MBARI:

On November 7, 2007 Todd, Kent and Kevin had a teleconference with Joe Van De Graaff and Jared from Alfresco. We had some questions about things and wanted to get a good feel for how Alfresco should be setup for the production environment. Joe first talked about the current situation with Alfresco licensing and where Alfresco sits with their non-profit status. Basically, Alfresco has changed from 50% discount to a 25% discount and they are not interested in name branding or references any more (more popular now). There are basically two version of Alfresco: Community and Enterprise. Community is the version we are using now and Enterprise is the only supported version. So the bottom line is that if we want any kind of support, we have to install Enterprise and to install Enterprise, we need a support contract. There is no way to install an unsupported Enterprise version. We then turned the discussion to Jared to talk about the technical aspects. I am not supremely confident that Jared has done any high visibility installations, so we may have to take a step of faith on some of his answers:

1. Do you have any general recommendations on server? Yes, see http://wiki.alfresco.com/wiki/Repository_Hardware. In general it looks like their configuration pages have improved see http://wiki.alfresco.com/wiki/Server_Configuration.
2. What database is best (knowing that we prefer SQL Server)? He said SQL Server is one of the four main supported servers (MySQL, SQL Server, PostgreSQL, and Oracle).
3. Should the database be colocated with Alfresco install? Should not matter unless you are going over REALLY long distances. Even if you install in same machine as Alfresco, there is no way to do direct communication, it will still go over TCP/IP stack.
4. Why are we having problems with CIFS on Mac? Hard to tell, but probably related to authenication. We should use passthru (we are). We asked about being on a LINUX box operating in a Windows Domain. He thought that might be related and running it on a Windows box might help.
5. Does Alfresco support Kerberos authentication? Yes. I can verify that they have documentation and examples of Kerberos setup, but I have not tried it. Todd was curious about this as this is where MBARI is probably going for single-sign-on authentication.
6. How do companies backup/restore Alfresco? His feeling was that the big companies are using Alfresco's export mechanism. This seems implausible to me as there is huge amount of overhead associated with exporting. He said that you could export only the rules and backup the content through CIFS. Again, this seems silly to me. The view through CIFS is not the file system level view of the files. Alfresco changes the files to manage them and then makes them look like native files when you view them through CIFS. By backing up through CIFS, you would have to restore through the Alfresco server and I am not sure that is even really possible. The export option seems better than that. It sound to me like the best way to backup Alfresco is to have the file storage on a NetAppliance and coordinate database backups with the NetAppliance backups. This would mean that if the server crashed, you could restore the filesystem and the DB to the correct state and point the new server to them. This should work, but timing is key. We could possibly have some Alfresco downtime when the backups occur, or just do the backups during a time when we feel like there would be little chance of changes occuring during backup. Not really an optimal situation.
7. How does failover work? There are two options: cold and hot. Hot is bascially clustering (you point both servers to the same DB and file location and the clustering keeps the local caches in synch), but then you have to pay for the CPU license for the hot backup machine. The other option (cold) is to have an identically configured server sitting dark, bug configured exactly the same way as the production server. This would allow you to make a DNS change and then fire up the cold server if you needed to do maintenance on the production server. The cold option means you don't need a separate license.

So, I think the bottom line is that we will want to have a dedicated Alfresco server, 1 CPU (quad core would be nice and maybe buy a machine with two sockets in case we want to upgrade later). Install Windows, point to a SQL Server instance that I.S. is maintaining, map a Tornado drive on the server and use that for the locale storage. Have an identical server configured but sitting dark. It will take some work, but setup a backup regime that will backup the NetAppliance share and the DB at the same time (this will need to be figured out).

Here is a diagram of what this might look like:

If this is where we want to go, the cheapest way might be the following:

1. Purchase a less potent server to use as the cold failover (maybe $4-5K)
2. Purchase Gold SLA contract with Alfresco at 25% discount ($11,250)
3. Configure cold server as alfresco.mbari.org, export content from oceana.shore.mbari.org and import to cold failover.
4. See if engineering (or I.S.?) has a server that is not being utilized that we could put in the DMZ and migrate the Confluence and JIRA applications to that server (call this server oceana.shore.mbari.org.
5. Reconfigure oceana.shore.mbari.org in the same way the cold failover was configured. Point to the existing DB and file stores, bring down the cold failover, make a DNS switch to point to alfresco.mbari.org. No export/import should be necessary.

So assuming we could find a server to host Confluence and JIRA, the bottom line cost would be around $15-16K with annual support contract of around $12K. At some later time, we would probably want to purchase a test server and license (total of about $8,000, with an additional $2,800 annual contract for the test installation).

Alfresco Deployment.jpg (image/jpeg)
Alfresco Deployment.jpg (image/jpeg)