Last year IBM purchased Platform Computing, I wrote a blog post at the time questioning how IBM would move forward in the HPC and cloud market. It has previously invested heavily in the cluster management system xCAT, but was acquiring a cluster manager, Platform HPC (PHPC) via Platform Computing. I wanted to know how IBM would combine these two investments?
The questions are still not fully answered, but I do understand that IBM is working on a product that will merge both xCAT and PHPC. While the developers work hard on what they do best, i.e. develop, I want to go through the pros and cons of the possible merger and share my only personal views on the different software elements that I would like to see kept or dropped!
As an avid user of xCAT, I will start from here. xCAT2, the latest version, comes from very old ‘IBM heritage’ and is the combination of xCAT1 and Cluster Systems Management (CSM) systems tools. Here’s a good overview of xCAT for any historians reading.
xCAT2 is one of the most dynamic provisioning systems available. It enables provisioning of various enterprise and open-source operating systems, as well as boasting stateful installations of Windows HPC server the ability to remember preceding events in a sequence of interactions with a user]. It also allows for multiple plugins for hardware and for all Intelligent Platform Management Interface flavours. There are even custom inputs for the HP ilo management engine and HP blade.
The biggest benefit of xCAT is that it is driven by perl plugins, so users can add new Operating Systems, hardware management and monitoring plugins with ease.
With all the fantastic features that xCAT does offer, it has one major drawback – the user learning curve. A user can do the same thing in multiple ways using xCAT; there is no right and wrong way. This extends the learning time.
Second, despite customers crying out loud for easy of use features, xCAT does not have a great user interface (GUI).
Platform HPC (PHPC)
By contrast to xCAT, with PHPC, the GUI drives everything. It is able to provision multiple packaged and imaged operating systems, using disk-full and diskless servers.
PHPC delivers a commercial grade scheduling system (using Platform Computing’s Load Scheduling Facility). Plus, an MPI suite (Platform MPI) that is used with many applications across commercial and research facilities.
In addition, PHPC includes the Platform Application Centre (PAC); which is a fantastic portal enabling easy job submission via a web GUI, including rich monitoring and statistics visuals. PAC enables administrators to create a customised web interface that makes life easy for users to submit jobs to the cluster [a simple matter of drag and drop to create the front-end], which in turn enables users to submit jobs with minimal clicks and with minimal knowledge of the underlying system.
The future match
PHPC’s PCM (Platform Cluster Manager) is the backend of PHPC that allows for provisioning of nodes. This is the portion of the whole solution that is likely to be replaced by xCAT. The main things that xCAT will bring to PHPC that PCM doesn’t have are
- Powerful command line interface
- Automatic discovery of nodes
- More management capabilities (including eventlogs, Vital Product Data and power information)
- To be able to manage more than 1000 nodes
- Support for a an extended range of O/S’s (including SLES, Fedora and Ubuntu)
- Installations using Internet Small Computer System Interface (ISCSI) and Statelite provisioning – a middle ground between stateful and stateless installations
- And many more features
Overall the features that the two will bring together will make it a fantastic product, so I am hoping this is achieved soon!