database performance: test setup: Rants
This is just a collection of complaints about problems we encountered during the setup and the tests.
FreeBSD
- Installation from CD continuously asks for the CD to be changed. Since the installer knows which CD each package is on, this unnecessary.
- The installer offers both bash v2 and v3 but both cannot be installed on the same system (at least not at install time) - so why does it allow the user to make this mistake?
- ports are installed in
/usr/local/bin
- but this is not on the path. - ports are very slow to update (Gentoo does a similar thing much quicker using rsync)
- Building java/jdk15 fails early without this (but it doesn't tell you when it is needed - when it breaks):
mount -t linprocfs linprocfs /compat/linux/proc
Surely, this could be detected earlier on. - Building jdk15 cannot be built with FreeBSD 6.2 (see these posts: 1 2
- The Linux JDK 1.5 port is for i686 only - no sign of the AMD64 JDK.
- FreeBSD cannot be installed on an extended DOS partition - which have been around for many many years
- Had to use binary packages from: freebsdfoundation.org to get Java
Solaris
- "Solaris Express Developer Edition" can only be installed from the DVD media, yet the CD installer allows you to select it at boot (and only fails later).
- After selecting DHCP for an interface, the installer still insisted that a valid IP address be specified for the DNS server...
- When customizing packages, the installer offers you the option of resolving the depencies at the end, but then selecting 'automatically resolve' did absolutely nothing! (had to resolve 134 dependencies by hand...)
- The disk setup wizard did not recognize standard partitions created with fdisk under Linux (FreeBSD and Windows have no problem with it)
- this is a known bug, due to the fact that the Linux swap partitions use the same partition id as Solaris.
(Even if the said partition is on an extended partition that the Solaris installer does not seem to recognize)
The proposed workaround involves restarting the installer and editing the partition using vi... No thanks. - Solaris failed to map a default PS/2 keyboard properly for the terminal, backspace came up as ^H
- The terminal is also incorrectly mapped for ssh usage, which means all the terminal codes look wrong (which leads to problems reading man pages for example)
SMF
has replacedinit.d
for most services (including ssh) which makes it really painful to restart services (since shell expansion will not findsvcadm restart network/ssh:default
).
Why not keep the init.d version as an alias to the "new" SMF method.- Solaris tar is broken, and has been for a very long time. It does not handle long filenames which can be very annoying for Java software which tends to use long package names and long filenames.
Fedora Core 6
- Fedora Core 6 hardly fits onto 2GB of disk space! More is less: it is now simply too bloated.
- Why do Fedora servers have pcmcia, bluetooth and ISDN services enabled by default?
- Dependency-hell: why do I need perl-DBD-MySQL when I just ask for MySQL to be installed?
- I would like to able to install just bash, init and openssh. This should not take more than 50 MB of disk space, including the man pages. The most basic install currently takes around 500MB!
Others
- Firebird 2.0.1 installation on Fedora Core 6 x86_64 did not work.
- Firebird passwords only have 8 significant digits!
- Sybase ASA sends broadcast packets on port 2638 to all network interfaces, without telling you in the installation notes - it broadcasts its identification through this port and will refuse to start if another instance on the subnet uses the same name.
- AbiWord's "table of contents" feature does not work, which made it harder to create the original PDF report
- glibc 2.4 and NPTL-only: I want my old kernel 2.4 back!
- Oracle installer cannot cope with IPv6 addresses for loopback!
- Why does the BIOS take so long to POST? I should really have installed LinuxBIOS... This test system takes 55 seconds to get to the bootloader!(grub).
- Postgres JDBC driver does not populate the error code and sql state when returning warnings, which makes it difficult to recognize what has happened
- and more importantly if this is fatal or not:
CREATE TABLE update_3569 (s2 varchar NOT NULL, s1 varchar NOT NULL, i2 int8 NOT NULL, i1 int8 NOT NULL, pk int4 UNIQUE NOT NULL, CONSTRAINT update_3569_pk_44 PRIMARY KEY (pk)) - warning[0]= class org.postgresql.util.PSQLWarning message=CREATE TABLE / PRIMARY KEY will create implicit index "update_3569_pk_44" for table "update_3569", sql state=00000, error code=0, severe=false