How NFS is unreliable for file-based locking

You hear a lot about how NFS is unreliable for file-based locking,but you rarely hear how and why, and understanding the details helpsin understanding what can go wrong. The fundamental source of NFS'sunreliability issue here is what I'll call the replay issue.

The communication between NFS clients and NFS servers is unreliable;both requests and replies can be dropped. Only the client worries aboutthis, and it uses a simple approach: if it didn't get an answer to itsrequest,


Weekly spam summary on June 9th, 2007

This week, we:

  • got 13,047 messages from 274 different IP addresses.
  • handled 19,786 sessions from 1,500 different IP addresses.
  • received 255,420 connections from at least 71,636 different IPaddresses.
  • hit a highwater of 12 connections being checked at once.

The volume is down compared to last week andprobably down overall, although not by much. The count of different IPaddresses is up a


Some thoughts on Slashdot's moderation system

Slashdot's comment moderation system is one of those wisdom of crowds thingsthat are periodically held up as exemplars; community feedback, allowing thebest to bubble to the top and so on. People who actually read Slashdotcomments, even only showing high scoring ones, may be aware that it doesn'tnecessarily work this way.

One of the reasons for this is that Slashdot's '+1 Funny' moderationoption tacitly encourages off-topic comments, because it encouragespeople to be funny even if there is


Why I hate firewalls, especially stateful firewalls

I hate firewalls because every firewall between two machines trying totalk to each other is another place for things to go wrong, which meansanother place to check (somehow) when things do go wrong.

Stateless firewalls at least have the grace to have consistent andpredictable behavior; if something is wrong, it is going to be wrong allthe time. Stateful firewalls make your life exciting by varying theirbehavior based on an ever-changing flux of generally unpredictable andinaccessible information, so things can go wrong now and right


Why you want a filesystem consistency checker

Filesystem consistency checkers have historically had three overlappingpurposes:

  1. to patch up the damage done when a machine was shut down partway through modifications to the filesystem.

  2. to find and fix up problems caused by corrupted disk blocks(whether they're caused by a dying disk, a controller error thatscribbled random data on a track, or whatever).

  3. to check for and repair structural errors created by operatingsystem bugs .

Fixing inconsistent filesystems is mostly or entirely obsolete thesedays, due to people moving to journaled filesystems that never


RPM's multi-architecture file ownership problem

An increasing number of Red Hat/Fedora machines are dual architecture:they run both 64-bit and 32-bit x86 programs, and thus need to haveshared libraries and RPMs and so on for both of them. This canexpose some interesting issues when installing and upgrading RPMs;one of them is the dual architecture file ownership problem.

There are a lot of RPMs that install both architecture-dependent fileslike binaries and shared libraries and architecture-independent fileslike header files and manpages. The


Why we need our SAN RAID controllers to support logical drives

Not all SAN RAID controllers support the idea of logical disks, whereyou can make a bunch of drives into a storage pool and then carve thestorage pool up into software-managed bits; some prefer the simplerand more straightforward approach of just exporting the pool itselfand staying out of the space management business.

Unfortunately, we need logical drives. The problem (and the reason) isthat logical drives are the only way to sensibly dynamically split thespace on a RAID controller between multiple fileservers. This is becauselogical


RAID-5 versus RAID-6

For those who haven't encountered it yet, RAID-6 is RAID-5 with twoparity blocks per stripe instead of one; it thus has a two disk overheadfor parity, compared to RAID-5's one disk.

We're interested in RAID-6 because it is more or less a better versionof RAID-5 plus a hot spare. With RAID-5, if you lose a drive the diskarray is non-redundant for the time it takes to rebuild onto your hot


Weekly spam summary on June 2nd, 2007

This week, we:

  • got 10,553 messages from 239 different IP addresses.
  • handled 19,451 sessions from 1,629 different IP addresses.
  • received 430,428 connections from at least 69,677 different IPaddresses.
  • hit a highwater of 9 connections being checked at once.

The connection volume is way up compared to last week ; however, I believe that it is mostly becauseof one machine, 213.


Why our Solaris fileservers still use the automounter

In theory, there's no need for your fileservers to use the automounter.Even if you make the filesystems visible under their normal names aswell as in /export , you could mount them explicitly in /etc/vfstab .However, doing so would significantly complicate our environment,because it turns out that the automounter worries about a number ofissues for us.

Of course, one big reason to use the automounter is administrativeconvenience; you can use the same set of automounter maps on yourfileservers as you do