FAQ
From OpenSSI
What is Single System Image (SSI) clustering?
At the most general level, the philosophy of SSI clustering is that the cluster looks like one very big machine, with applications and management working like they do on a single machine. SSI clusters usually have one root file system, so that the system operates in a cohesive working enviornment.
What advantages does SSI have over alternative forms of clustering?
Manageability. An SSI cluster can be managed like a giant SMP machine.
What is OpenSSI?
OpenSSI is an open source clustering project that addresses three major issues that cut across all the various cluster segments: high availability, scalability, and manageability. It intends to become the definitive project that unites all the Linux cluster factions.
What advantages does OpenSSI offer over other Single System Image clusters?
Manageability, availability, scalability and usability.
Unlike other Single System Image software, OpenSSI operates with one main root filesystem which runs on one specific node, allowing other nodes to be added and removed from the cluster without disrupting the cluster environment. The single root filesystem allows changes to be made from any node and be immediately visible to the entire cluster, without the need for replication. It also allows for simple integration of additional nodes with almost any sort of hardware configuration.
Although having the root filesystem on one specific node is a potential single-point-of-failure, root failover is available. OpenSSI has high-availability mechanisms for networking, filesystem and applications.
The scalability advantage is not in just the number of nodes one can aggregate, but also the resource-sharing between processes on different nodes, allowing one to easily scale many applications.
When will OpenSSI have full support for Linux 2.6?
OpenSSI 1.9.0 release made on 27th April 2005 is based on the Linux 2.6 kernel. The latest release supporting 2.6 kernel is 1.9.2.
How does OpenSSI integrate with LSF?
Currently, LSF integration has been tried to a limited extent. LSF works fine on on a 2-node OpenSSI cluster. Currently, we are in the process of trying it on more than 2 nodes. On a related note, OpenSSI has been integrated with SLURM and OpenPBS as well.
How does OpenSSI compare with other clustering solutions?
The following contain some comparisons:
- The paper from INRIA on "OpenMosix, OpenSSI and Kerrighed: A comparitive study" available here. Bruce J. Walker, the OpenSSI project manager, commented on this paper here.
- Bruce J. Walker's paper on A Comparison of OpenSSI and OpenMosix”. Comments on this comparison from the OpenMosix folks is available here.
How does OpenSSI stand up to database I/O benchmarking?
We are yet to run any database benchmarks (TPC-C, TPC-D and similar) on OpenSSI clusters.
Does OpenSSI support the PostgreSQL database?
OpenSSI has been integrated and tested with both MySQL as well as Oracle 9i RAC. We are yet to integrate and test with PostgreSQL. Significant performance bottlenecks but be the case as PostgreSQL is based on spinlocks and shared memory. Although not tested extensively yet we expect some performance bottleneck.
Here's a doc that gives step-by-step instructions on how to install PostgreSQL on OpenSSI enabled Knoppix distro.
Can OpenSSI do triggered process migration?
We have users that want to exploit OpenSSI in compute-intensive application environments, where batch jobs run on different architectures and nodes integrated with LSF. Those users need a way of saying "migrate these jobs under these (load) conditions", where the optimal distribution of jobs is not known a priori.
This is possible with some restrictions. LSF is already integrated with OpenSSI. So are SLURM and OpenPBS. The administrator can configure process migration to either happen in an automated fashion under the control of OpenSSI or under the control of the running application, via interfaces provided by OpenSSI. Both command-line as well as programmatic interfaces for process migration are provided in OpenSSI. If you have not already compiled the application for different architectures (which is most often the case), then process migration may have to be restricted to a set of same-architecture nodes. The application needs to ensure that. One way to do this would be to group architecturally similar nodes into distinct OpenSSI Clusters, and use LSF operate on those clusters instead of each individual node in a cluster.
Can OpenSSI fail over active TCP connections?
Once a TCP connection is established with an OpenSSI node, the failure of that node will not cause the connection to get failed over to another node. Implementing such a system is essentially impossible, and not within the scope of HA-LVS. However, assuming you have HA-LVS properly configured, the failed node will transparently be removed from the real server pool, and no further connections will be handed to it by the director node until it is brought back up.
Does OpenSSI cluster filesystem (CFS) support swapfiles?
Currently OpenSSI does not support swapfiles on CFS. We support physical swap partitions.
