Daring to be the same

Today was different. A customer came to see us, which in my job is rare, too rare. The high spot for me was watching a colleague do his Dtrace demo, which is highly interactive with the audience, not on his laptop, but on our production Sun Ray server. None of the users noticed even when he traced every call to mutex_enter, which even as it was being typed I did wonder if it was wise.

The other thing to come out was the question of when to be different. The customer has a highly customised environment with minimal packages installed and no “unneeded” daemons running. Which is fine, there are good reasons to do this, but I just had to ask, why? What was the business reason for not running each daemon and not installing each package? Well it was to reduce risk, risk of security breach and reduce the amount of stuff they had to support. However in reducing risk they had opened up a number of problems with the system, it ran, but there were problems with a number of subsystems that would not have been seen had the system been left alone. What in my opinion was missing from the risk assessment was the risk of being different. By heavily customising the systems to a minimal system they became so different from any other system that they are almost guaranteed to find problems that are not seen by anyone else.

It is not to say that this should not work in an ideal world, but this is not an ideal world. Running large systems is about risk management and reducing the unexpected. I hope that the customer went away with a better understanding of the risks that are exposed by being different. If being different does not get you an edge over your competitors then dare to be the same.

Comments:

You talk about the risk of heavily customizing a system, but in my view much of the problem is due to the way Solaris is packaged. I don't know anyone who doesn't heavily customize the installation. And why is this? The default Solaris installation and clusters are practically useless in the real world, so that you have to customize, and everyone does it differently. Solaris ought to be better packaged, and split into packages that are both many fewer and meaningful to customers, and the diversity of installed systems would be much reduced.

Posted by Peter Tribble on February 18, 2005 at 11:11 AM GMT #

I'm not sure that I would agree that the default are practically useless in the "real world". I live in the real world and use them all the time. The added "benefit" of not having the window system on a server is small compared to the risk associated with having the bits there.

Perhaps things will be improved with 10 now that there are package catabories. Though I do not in any way dispute Solaris could be better packaged or that the package tools could do a better job, but in the current state people should understand that there is a risk to being different that has to be factored into a decision to customize.

Posted by Chris Gerhard on February 21, 2005 at 07:36 AM GMT #

We run into a problem during patch time. There are so many packages on the system and we have to patch them all. Even when we know the subsystem is not being used. Case in point, why do we need to patch the USB subsystem on an E220R (it does not have USB at all), or the audio subsystem on a mail server? Why should we be running programs that are not part of the machines job? This is just inviting problems. Most of our systems do not use kerberos, but when the advisory came out, we had to patch all of them, because it was installed. That was quite a waste of sysadm time to take care of.

Posted by Jim Pirzyk on April 26, 2005 at 02:32 PM BST #

Agreed. The Solaris installation packages are not aligned with many administration needs. The core package doesn't install without errors and is missing a LOT of useful software. Apparently Sun believes sysadm's are off their rocker for attempting to drop X from a syslog machine, a web server, a... I'm hoping OpenBSD folks will lend a hand with OpenSolaris and teach a few deaf Sun employees about building UP systems, small foot prints, etc. (Ok, its a pipe dream). Back to Google to learn more about building up from core... :-(

Posted by Robert Miller on June 14, 2005 at 08:24 AM BST #

I would have to agree. The solaris packaging system is odd. OpenBSD got it right.

Posted by vext01 on May 10, 2006 at 09:04 AM BST #

Post a Comment:
Comments are closed for this entry.
About

This is the old blog of Chris Gerhard. It has mostly moved to http://chrisgerhard.wordpress.com

Search

Archives
« April 2014
MonTueWedThuFriSatSun
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
    
       
Today