Re: Re: Usage of tools

[dws]

That's an argument I do not buy. I've worked too long in sysadmin departments. Maintainance of tools is important (and needed). And therefore, desktops shall have NO tools. Nada. You put them on a central file server, and use NFS (or whatever they use in a Windows environment) to install something once, and make it available everywhere.

Our experiences differ here. I've seldom worked at a place where run-everything-off-one-server was a stable solution, usually because of a geographic split between servers and developers, complicated by building moves and "oh f*ck, somebody didn't arrange for enough bandwidth between building A and B."

A run-everything-off-one-server configuration also introduces a single point of failure. You can protect the server with RAID5 and big UPSs, but it's still a single box sitting there waiting for the gremlins to sink their teeth into it. If a single developer box goes out, replacing it is easy, and at worse you lose a developer day. If the server everyone depends on goes out, you're hosed. And for some strange reason, hosing always seems to happen about 12 hours before a major deadline.

[Abigail-II]

Let's see, you don't want a central file server because that introduces a single point of failure? Then what are you going to do with your data and products? Store the source code of your product scattered over hundreds of PC's? Not use a database server at all? Let me tell you, you will lose far more than one workday if a PC goes down if you do it that way.

The "single point of failure" has been a known issue for a long, long time. And it has been solved for a long, long time: you'd use a cluster. Many vendors write cluster software, and it works very well in practise. File servers are really easy to cluster, you can have one up and running in less than a day.

Abigail

[dws]

Then what are you going to do with your data and products? Store the source code of your product scattered over hundreds of PC's? Not use a database server at all?

Not at all. We "flew under radar" and installed our own CVS server local to the group. The IT folks, who we depended on for things like adequate bandwidth between buildings (including the data center), didn't know about it. We would batch up our changes and update the official server periodically. And we had plenty of local "database" (RDBMS) servers for development, though I suspect that's not what you meant.

The problem is that the same people we would be counting on to set up a cluster for us are the same ones who couldn't set up a reliable pipe between our group (which had been moved to a "new" building), and the corporate data center. They had their hands full. We couldn't wait 6 months, so we worked around the situation with some extra hardward that we had available in the group. Nobody lost any work.

Update: Sparring aside, I think we're in agreement. A sensible organization supports development by setting up a reliable server cluster. But most my work has been in small, growing companies, who experience growing pains that make them non-sensible, and that have to be worked around.