winscape 2371

This may be the next big thing in open source.

Although it may be easier to address it at a protocol-service layer instead.

OpenBSD has seen some of this lately, where a separate box tracks and keeps all state and can take over in a few hundred milliseconds.

This is an issue everyone agrees upon. There are two separate tug-of-war games going on. Gadget manufactureres mostly hasn't realized how important drivers are for their livelihood. They peddle rudimentary drivers to the MS world, and MS balks due to the instability they introduce into windows. They have tried carrot with educational programs and partner stuff for a few years, but it hasn't been effective. I expect we will soon see the stick.

They are trying to push the same shoddy stuff on Linux-BSD developers, and are peddling "binary interface layers" etc. Linux developers are not impressed. They have seen the drivers, and have fixed a few thousand of them to be workable. The drivers coming out of the likes of Via, SMC, acer etc. are at best considered early drafts of what the driver is supposed to look like. Others are normally worse, and the driver has to be reimplemented from scratch.

My thoughts are a little different. I accept that individual hardware has downtime. As any technical device it has to be services. I just want it to 1) fail in a controlled and graceful manner 2) be easy to restore 3) services provided by that hardware to easily be shifted to other hardware.

ECC, raids multiple independent processors and buses help with 1), but to acheive 3) I need a redundant network anyway.

A sysadmin I admire, but cannot name, once stated that he planned to have a whole building blown up without losing a single, committed transaction. For this business I am aiming at that.

The stuff we need slowly trickle into Linux-BSD stuff. The raid and ECC failure stuff is quite good now. Linux 2.6 is a quantum leap forward. Several journalling file systems, good hardware and software raid support, the "master panic" is replaced by attempts to resque the system etc.

But, you get what you pay for. There is a place for heavy Sun-SGI-IBM gear as well.

-- mrr