It is not really the problem with Linux OS itself, which meant to be of humble and people oriented, but problem stems out from the philosophy of distribution designers who want to dominate the users with unwated technologies aimed to make Linux par with MS OS such as Logical Volume Set that I care less about. I have already built i960 or i80303 assisted LVM using ICP Vortex SCSI HBA without using any CPU resource back in 1990s. Hence I do not want something tapping out CPU resource being a humble purist.
1. Idealistically a Linux installer should have an option where user can launch an xterm and run fdisk, mkef3fs, mkswap and edit proposed mtab before installation starts! It could be run from a transparent ram disk. At minimum, Linux should honour user pre-defined partitions with mounting points and no question asked.
This will solve all my problems I have experienced with recent Linux install. In 1990ís nearly all Linux distributions used ďutil-linux-fdiskĒ and user specified mounting points therefore user had more chices to build own layout similar with other Unix variants. Why not give that option again? It does not have to be an xterm, it can be like simple disk layout program built into Solaris 10 install. Solaris does not probe other partitions like current Linux does and that is definitely better for some users. In the past, Solaris did all kind of damages to existing partitions but it has learnt to happily coexist with other operating system today. Solaris is becoming one of most powerful but also people oriented operating system all over the world.
In fact I have recently bought Fedora 9 and open SUSE Linux11 and was not happy with the outcome at least one of them did to my disks. One of them realigned chain order on extended partition before even I committed to install. Later a close inspection revealed that it made some effort to consolidate all preformatted ext3 partitions to form a LVM without even asking me. What a disaster! I consider it a serious misconduct. You just donít go to your neighbours house and paint it over to different colour without permission. I will now have to spend entire weekend to rebuild the disk back to the way it was with clean chaining. Unlike primary partitions, it involves manual editing of 11 EPBR sector pages 512byte each on the disk. Delete and reformat partitions will not bring the exact chaining order as it was. I have already spent hours for partition planning and create partitions in a clean consecutive order and Linux should have honoured predefined and/or preformatted ext2/3 partitions and not to change its logical chaining order or realign. I have tools to view and edit any sector up to 300GB and I chain them manually. After this incident I could not trust this installation program any farther so I could not install it. I am not happy unless I know CHS geometry for each partition and able to make snapshot file for recovery. LVM is out of question. I do not like it. I want to stick with very basic. It helps if partitions are daisy chained in an order relative to sector number. This misfortune installation attempt happened to be on my first SATA platform without my most familiar presence of SCSI due to space factor. I have used SCSI and ESDI (up to 1995) but have never attempted to use ATA/IDE to this date. I understood and extensively used the true SCSI LVM system consists of I/O bridged RISC CPU like i960 or 803XX can create in ICP Vortex subsystem. Hence Linux OS should not even play any game to mimic the LVM which I knew into register level already from 1990ís technology.
2. Kernel needs block device access file system reworked to be par with todayís disk size!
I had preformatted ext3 and swap partitions and was willing to give the installer programme mounting points, hence Linux should not tell user what to do about other partitions. Linux has itís own limitations that I am aware of: so called 15 partition rule per disk. Sometimes itís 12 partition maximum when there is no primary partition exists. Terminology of 15 partitions maximum is very ambiguous. It seems to me that it is pre-defined in current kernel, sda sda1 sda2 . . . . . sda15 as device files. Therefore, if I create one extended partition taking entire disk space with 12 logical partitions in that extended partition, Kernel still sees as 16 partitions even though actuality is 13 since device space between sda2 and sda4 are not used. This scheme needs to be urgently updated to meet todayís disk space. Giving one more bit gives you 32 device space per disk that is more realistic for todayís disk size. I know this sda sda1. . .sda15 (sda without number being MBR/partition table and being one device) scheme go back to late 1980s and no one has made any effort to update it. It is not adequate for the age of 1 terabyte disks. I have two laptop sized miniature 300GB disks (2.5Ē disk runs at 10000RPM with read speed exceeding 135MB/sec) to try out to compare with other existing SCSI systems.
Note: My Opinion regarding LVM.
My kind of LVM is totally transparent to OS and OS does not even have to know LVM. For a simple example, with ICP Vortex GDT8643RZ, you can create 7 disks (LVM) 40GB each out of two140GB Cheetahs. Or a single 560GB disk (LVM) out of 4 Cheetahs. Translation is done between firmware, I/O processor and dedicated 256 MB memory. Linux nor Solaris do not have to know what is behind PCI-X64 bridge on HBA. This hardware based LVM strictly sticks to extended int13 scheme and never have had any problem with it. Only the difference is that you can not resize int 13 hooked LVM without imaging each LVM onto off chain conventional SCSI disks on Adaptec 39160 (yes I hate 39320 because of driver has feature of software based LVM) because int13 hooked LVM are not DLVM. But you can make partitions active or hide or multiboot without LILO or GRUB using system commander or OS/2 Boot manager. ICP Subsystem also supported from legacy OSes DOS. Win3.1, Win9X, Win NT, Win XP, OS/2 to cutting edge Solaris, SGI, Linux and like as late as 2003 with tremendous performance gain. However the space and heat factor requiring of cooling fans with bulky cabling and manual jumper setup elsewhere killed these ultimate SCSI subsystem that I trust. I will perhaps continue using them until my death due to their trust worthiness and amount of spare disks that I own. They are suitable for large TIF Photo file editing and also for continues 3D rending of topographic data with building objects and vegetation data turned on without knowing fatigue seeking random data as well as MRI and CAT Scan.
SCSI LVM only comes slightly behind fibre channels but by far beyond any OS created LVM under Linux and Windows OS. I am a purist. I do not want main CPU between chipset to mimic LVM, audio card and I/O processor using software no matter how fast CPU gets. I enjoy distributed processing that let you format your Canon memory card whilst downloading a file via internet and printing a photo file (processor sees as if just a memory dump) because the printer has i960 CPU of itís own. The presence of software based LVM, multi-tasking will not be transparent. Acrobat viewer will no longer be of a snap. If the system has to perform paging in Dynamic Logical Volume Manager also pays 12 % plus penalty in speed.