At work we have been toying around with using a UNC content store to host our applications. The idea is that we can have n front end web servers and 1 centralized content store for app files, content and images. This gets us around the current limitation of Application Center 2000, which can only support 12 servers with a replication size of about 500MB. Thus far we have had some serious issues to say the least.
Our UNC content store is running out of work items. This seems to be related to SMB and the amount of ram available to the non paged pool. The non paged pool limit is set around 250MB and isn't tweakable. Each work item equates to around 20K. The maximum amount of work items we can have at one time is around 12,000-13,000. We have heard several large companies like Walmart and Ford Motor Company are using this configuration, but I think our application is architected in a completely different manner.
In our application we are hosting several thousand virtual websites for automotive dealerships. We have around 1,500 active websites all running from the same codebase. When we attempt to move this configuration into production our UNC content store lasts about 15-20 minutes and then goes completely kaput.
We have a PSS call in right now, so hopefully I can elaborate on a solution once we find it. I was wondering if anyone else has used a similar configuration for their high availability applications.