Adam Tauno Williams said the following on 05/10/2013 10:15 AM:
If you don't like openSuse with systemd ... Fortunately I do. 20+ years of UNIX administration ... die SysV die. I am so tired of sermons from those of the tired religion of the holy text file. If someones vision of good system administration involves grep and regular expressions I don't want to be employed beside them. These things are HACKS, are always broken (or about the break), and very difficult to figure out when you go into a site.
I too like systemd and the direction it is taking us. UNIX - and Linux - is a real computer system, not a toy. The big server farms and the array servers are demonstrating that. I'm glad I used V7 and all the small systems; they were a great learning experience, but I HATE HATE HATE what USG did. Lots of bad code and it set the way for breaking the clarity and simplcity of Dennis's creation. Yo me, USG let Entropy in. Looking back, I can see that as a necessary step in the commercialization, but as it documented a lot of stuff, but it lead to things like the X-Wars and the media hasn't given up yet on the idea that the the diversity of *NIX is an advantage over the totalitarianism of the 'locked in' systems. Scripting is great, until it became unmanageable; and the SysVinit suite has long since become a moras. I think it was Gerry Weinberg who coined the idea of "5 finger code", referring to using your fingers to keep marks as you clip back and forth in the 14" fanfold listing or spaghetti code of the COBOL and Assembler days. SysVinit has become like that. Dennis's v7 had simple, clear "do one thing and do it well" scripts, and the SyVini scripts are a complete counterpoint to that clarity. As a home user neiother I not Linda can eb viewed as the main market for UNIX and Linux. I have clients who need systems that aren't difficult to figure out, and the ways systemd works fits that. The way systemd logs is great for them. Even now, well ... Under my desk is one of the crapped out 800Mhz box from the Closet of Anxieties. Its running a DNS server for the other such boxes. The DNS loads a huge set from http://pgl.yoyo.org/as/serverlist.php This morging I watched it load and saw ... Feb 8 10:04:33 server named-checkconf[1810]: zone ad2.lupa.cz/IN: loaded serial 2012100600 Feb 8 10:04:33 server named-checkconf[1810]: zone ad2.mamma.com/IN: loaded serial 2012100600 Feb 8 10:04:33 server rsyslogd-2177: imuxsock begins to drop messages from pid 1810 due to rate-limiting Feb 8 10:04:35 server rsyslogd-2177: imuxsock lost 268 messages from pid 1810 due to rate-limiting Even this crappy old machine can outpace the syslog daemon writing to a text file. Now some of my clients are using syslog to monitor networks of hundreds of machines. And in order to mine the data they are sticking it into a database so they can perform queries on it, ones that are more involved than you can do with grep and regular expressions. Text files can't cut it any more. Even back in the 80s many university sites found that their machines ground to a halt when someone logged in. One of the assumptions that Denis made for the kernel was that the kernel tables were small and so didn't need complicated algorithms. The overhead of things like binary injection/search was little advantage over linear when there were so few entries. But that didn't hold when your password file had thousands of entries, the universities found. So even at BSD4.1/4.2 there were patches for password files that were indexed. Now we have corporations where *NIX is handling the mail for hundreds of thousands of employees. The overhead of accessing a SQL database (which is what's behind many LDAP servers) is trivial by comparison. Heck, even back in 1990 I was using RADIUS at an ISP. Text files can't cut it any more. Well, that's not an ABSOLUTE. Those examples are about high volumes that need to be accessed fast and randomly. (Rather like a disk, eh? So use the BtrFS!) Systemd is, in effect, the system version of what inetd is for network daemons -- a dispatcher and manager. Yes, they could both have their config files in a SQL style database but pretty soon you'll have the Windows Registry. That's neither needed not useful in this context. We're not talking about a high volume or entries or a high rate of repeated access with inetd/xinetd or systemd. Perhaps the debate should be about human intelligible vs machine optimized text files. The style used in systemd, xinit, xorg.conf are documented externally but allows for in-line comments, something I sorely miss when dealing with things like Windows registry. The XML format has the semantics documented elsewhere, hopefully, with the DTD, but how much semantic for the human is there in that? Human manageable configuration text files have their place. Perhaps, just as in _some_ systems, the password file has evolved into something indexed and shared so might systemd and so might xinetd and so might xorg.conf, but I don't see it for any of them in the immediate future. There really isn't the pressure as there is with the high volume or high transaction rate systems I mentioned to open with. -- People who won't quit making the same mistake over and over are what we call conservatives. - Richard Ford, in his novel Independence Day -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org