*BSD News Article 21368


Return to BSD News archive

Xref: sserve comp.unix.bsd:12671 comp.unix.wizards:30890 comp.sys.sun.apps:4869 comp.sys.sun.misc:9682
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!agate!CS.Berkeley.EDU!eric
From: eric@CS.Berkeley.EDU (Eric Allman)
Newsgroups: comp.unix.bsd,comp.unix.wizards,comp.sys.sun.apps,comp.sys.sun.misc
Subject: Re: checkpointing software for UNIX or SunOS
Date: 23 Sep 1993 14:41:57 GMT
Organization: UC Berkeley Mammoth Project
Lines: 19
Sender: eric@mastodon.CS.Berkeley.EDU (Eric Allman)
Distribution: world
Message-ID: <27scjl$qjc@agate.berkeley.edu>
References: <TMB.93Sep23012442@arolla.idiap.ch> <VIXIE.93Sep22202934@gw.home.vix.com>
Reply-To: eric@CS.Berkeley.EDU
NNTP-Posting-Host: mastodon.cs.berkeley.edu

In article <VIXIE.93Sep22202934@gw.home.vix.com>, vixie@gw.home.vix.com (Paul A Vixie) writes:
|> I've seen several "undump"-like facilities built into shells and kernels to
|> provide checkpointing, but the user community that can benefit from them is
|> small and getting smaller since almost all long-running processes have 
|> network connections open to various servers (even implicit ones such as a
|> connection to kerberos or hesiod or DNS), or they have files open to system
|> configuration files (/etc/services, /etc/passwd) which might not be the same
|> size and layout after a restart.  
|> 
|> Processes that need checkpointing generally have to implement it themselves,
|> as Sendmail does with its frozen configuration file and as GNU Emacs does 
|> with its "undump" mechanism.

It's worth pointing out that this is VERY hard to get right and even
harder to do portably.  For example, there can be system routines that
have resources open that you know nothing about.  I've taking the freeze
file out of sendmail entirely -- it was more trouble than it was worth.

eric