I don't think thousands of forests on one host will be practical. Just to pick one problem, each forest writes to its label once per second. So thousands of forests will drive thousands of writes/sec, without even running any queries.
So I'd rethink the physical isolation: document permissions are the right technical solution. Use prefixes to provide unique usernames. -- Mike > On Dec 18, 2014, at 13:50, Gary Russo <[email protected]> wrote: > > I’m building a MarkLogic cluster that will be used by accounting firms to > manage client tax data. > > Notes: > · Each accounting firm will have a unique set of users and clients. > · Every accounting firm needs their own isolated “dataspace” to store > client data. > · Need to support 5,000 accounting firms. > · Need to have a “physical” firewall to isolate data of each accounting > firm. > · It is very important to not commingle data between accounting firms. > > Example: > · Account Firm A – Will have 20 users that will manage data for 40 > clients. > · Account Firm B – Will have 25 users and will manage data for 100 > clients. > · Account Firm C – Will have 100 users and will manage data for 500 > clients. > > > I see 2 approaches: option 1 - Application/User Level and option 2 - > Port/Database Level > > Option 1 is the salesforce.com approach where each firm has a unique REST > endpoint with unique set of users/permissions. > > Option 2 is to give every accounting firm a unique database and port numbers. > This means the MarkLogic cluster will have more than ~10,000 forests with > ~5,000 unique HTTP servers. > > If Option 2 is used, the accounting firm on boarding process will be fully > automated. A web app will be created that will utilize the REST Management > APIs. => http://docs.marklogic.com/REST/management > > > My questions: > 1. What is the best approach to support the “multiple tenants” like > this? > 2. For option 2, is the use of unique databases/port numbers for 5,000 > firms considered too much? > 3. For option2, how will 10,000 forests on a 3 node cluster impact > performance considering the 2 CPU core per forest rule of thumb? > > > > Gary Russo > Enterprise NoSQL Developer > Phone: 201-536-4432 > Skype: garyprusso > http://garyrusso.wordpress.com > > _______________________________________________ > General mailing list > [email protected] > http://developer.marklogic.com/mailman/listinfo/general
_______________________________________________ General mailing list [email protected] http://developer.marklogic.com/mailman/listinfo/general
