[DC LoCo] Upgrade from 10.10 -> 11.04 took two days to recover from

[Ken Stailey]

--- On Sat, 5/28/11, Robert Simmons <rsimmons0 at gmail.com> wrote:

> What is the purpose of the box that the OP was talking
> about? 

It's a home desktop in a house full of desktops and laptops.

Admittedly I consider it to be the primary desktop which is why I upgraded it late in the game after building a number of 11.04 VM guests and test systems, none of which were using DMRAID ("fakeraid") and even if they were I don't have another PDC RAID controller, only NVIDIA or such on other systems.

None of the systems that we have at work can even be called "production" because of political reasons.  I have taken to calling them "live systems" to get around that.  We are a dev shop and have a lot of leniency but don't like to take advantage of that in ways that would interfere with real developer work; we do maintenance work that requires downtime after-hours on a routine schedule that developers can plan around.

At work we are still running some non-LTS systems but the reason is that back in the bad old days of a year or two ago I didn't know enough about how to build custom deb packages like I do now.  There were major deficiencies in 8.04 LTS which we surmounted by using point releases.  It is a mistake we have almost completely recovered from.  It is somewhat amusing to think that we still have one 8.10 NAS system left that is pinned behind a purchase of right-angle power cords for its replacement.  

Of note is the fact that the Ubuntu 10.04 LTS Linux version 2.6.32 is amazingly slower than 2.6.38 on SMP systems.  I have been deploying Linux 2.6.38 on our "live" systems and each time the load average drops phenomenally.  The reason is that in Linux 2.0 SMP was added with absolutely no ability to run kernel code on more than one processor concurrently.  Linux 2.2 introduced the BKL to facilitate transition to a true SMP kernel but the BKL was such a compromise still.  Starting with 2.6.37 and ending with 2.6.39 removal of the BKL accelerated and was completed.  Even the partially-removed BKL Linux 2.6.38 is vastly superior at executing kernel code simultaneously on multiple processors (or processor cores) than Linux 2.6.32 is.