|Active/Active architecture||Frank||2/19/12 10:57 AM|
IG is called a distributed DB. We looked at the documentation, but cannot find any information on the support of the following architecture:
- IG db is configured in an active/active configuration across 2 or more nodes (i.e. physical server)
- Each node runs an identical copy of the database, and changes to one database are replicated in real time to the other nodes
- Optional : support for load balancing of db requests across the existing nodes
We are trying to architect our solution to achieve High availability while maintaining very good scalability to support a very high number of concurrent clients, but are not sure if IG is a good fit.
Your advise is highly appreciated
|Re: Active/Active architecture||Asher Moody-Davis||2/21/12 8:21 AM|
Each graph is accessed using a properties file. This file defines the
location of the bootfile, storage locations, and configuration
information. This makes it easy to define multiple graphs and connect
to them separately and independently. InfiniteGraph also allows you
to create multiple storage locations for a single graph instance.
This means that portions of the graph can be stored on multiple
machines across the network.
InfiniteGraph is implemented using MROW which allows concurrent
reading of the graph but only one writer to a particular section of
the graph at a time. Locking contention can be improved by limiting
write transactions and configuring your container settings in the
Let me know if this brings up more questions.
|Re: Active/Active architecture||Frank||2/21/12 11:04 AM|
Thanks Asher. Yes, it does:
1. Let's assume I have a graph X in db D. D should be highly available.
2. How do I setup two (or more dbs) D and D' in HA mode. How is data written to D automatically synchronized with D' ?
3. What's your architecture behind this : multi-master , master/many slaves?
4. If this is not yet available, what is your roadmap.
Where do i find the respective documentation on this.
Thanks in advance
|Re: Active/Active architecture||Asher Moody-Davis||2/22/12 3:55 PM|
InfiniteGraph is distributed but does not currently support high
availability. Currently, the user is responsible for any replication
of data. However, we do plan to offer replication with eventual
consistency in the future. The goal is to offer consistency,
availability, and partition tolerance (CAP).