Emailed to the following IBM and USC people on Thu Aug 29, 2002: David L Nemiroff , Dwain Bitter , Bruce M Potter , Dave Jursik , Dewey Dasher , Peter Ungaro , Asbed Bedrossian , Jim Pepin , Garrick Staples , Dennis Smith , and Egan Ford . You said you appreciated candor, so here's the collection of my concerns after the conference call today: I'm presented with one reason why I shouldn't use xcat; it lacks support for the new RSA cards. This leads me to wonder how CSM can support it but xcat doesn't. Does CSM actually support the new RSA cards today? 1) I'm not convinced that xcat can't support these cards, or 2) this sounds like a hardware problem. CSM doesn't use flat files to store the node data. Flat files are always easier. They can be quickly built and maintained with simple Unix commands like sed, awk, and seq. I don't know how I'll be able to interact with CSM's binary database. According to the online CSM redbook, which reads like it's out of date (but has a recent publishing date), CSM doesn't support service processer programming and remote BIOS flashing. The redbook instructs me to actually boot each node with a floppy and manually configure each processor and management adaptor. In the time I save with those xcat features, I could probably extend it to work with the new RSA. MAC address collection? dhcpd and named conf file generation? These are all important concepts to xcat that I couldn't find in the redbook. Remote video? I can't find any mention of remote video in CSM. Noderange? I can't find any mention of an equivalent to xcat's powerful noderange function. While it seems minor, it's actually a central feature. The design of noderange has been a reoccuring topic on the xcat mailling list. ELS configuration? GM mapping handling? Effective ssh host key management (host keys need to survive node reinstalls)? Build scripts for mpich, mpich-gm, conserver, pbs, maui, gm, etc...? Are these all missing features? Does CSM have at least a quarter of xcat's features? CSM will not have xcat's user community. I'm not sure IBM realizes how much I value xcat's user community. Unfortunately, the closed nature of CSM will never stimulate such communication. Egan. Where is Egan in all this? He has done so much for all of us. That guy's knowledge is invaluable. He is a significant reason why I've wanted to stay with IBM so far. Scalability and extensibility. It's hard to imagine an implementation that will scale and extend better than xcat. Of course, having never seen CSM, I can't comment on it. But IBM's recent trends with regards to the service processors and mpcli have been discouraging in these areas. (I'm addressing the hardware sp problem elsewhere) The event reporting software is not compelling. I've already written my own, twice. And there are any number of open source alternatives today. xcat is already including "mon" which I've not tried yet, but I hear does great things. In conclusion, This is showing a trend that one day I'll get wiz-bang GUIs instead of workable CLI utilities. This troubles me. Does IBM know that I require CLIs instead of GUIs? Is "require" a strong enough word? Perhaps "mandate" is a better word? There is no "admin desktop" on my cluster; X isn't even installed. Jim Pepin noted the same trend when he was refering to IBM's catering to Windows and abandoning Unix/Linux users. It is the open environment around xcat that is my personal compelling reason to stay with IBM. I watch how USC interacts with other closed-source vendors (Oracle, Clarify, Blackboard, SUN, etc) and cringe at the thought of having to deal with those models. Why not GPL xcat? Let the xcat community live a life of its own. I can dream up any number of applications that IBM can develop and sell to additionally support xcat. Also, many companies are using a hybrid model to let their advanced users run in an open environment (constant updates via CVS or similiar) and selling finished products on a standard release cycle. I understand that IBM needs a scalable method to support cluster management, but I don't see why xcat needs to be squelched in the process. In this user's opinion, open standards, protocols, and software are always better then their closed and proprietary counterparts. I already have the open xcat that does everything I want. How am I to be sold on something that is closed and doesn't do everything I want?