AFAIK, the BIOS only knows about the four primary partitions. It cannot access Solaris slices, BSD partitions nor extended partitions.
Extended partitions and Solaris slices are different and uncompatible ways of subdividing a primary partition in smaller parts. It is true that Solaris requires a primary partition to be installed.
Linux with UFS support can mount Solaris slices with no problem. The boundary issue is unrelated. Each partioning program has its own logic about where cylinder boundaries are. Linux and Solaris can indeed disagree but it's a minor issue that should just be ignored. Anyway, cylinders are no more a reality with modern hard disks.

These look like good idea's.

As I installed my system on top of an LVM on my server I was able to resize and make a new /var partition on a much faster disk.

Just wanted to mention, using noexec on /var borks your apt-get.

jlliagre,

I am also of the opinion that the objection raised by Linux in an extended partition with a Solaris present may be harmless as you suggested. However it is something Linux cannot cope as it is reported in various partitioning programs.

My whole point of the post is a MS system calls a drive F and that can be any partition and in any disk. MS system has only one extended partition.

Linux is trying to be compatiable. Grub can identify the 4th partition of 2nd disk by a notation of (hd1,3).

Solaris goes one further with (hd1,4,a). Thus Unix-biased system can have sub-slices in "every" primary partition in a hard disk.

Also Solaris and BSD systems seem to have an issue if there are too many logical partitions inside an extended partition because that is some thing these systems may be functionally uncompatible. I use the maximum the Linux limit of 59 logical partitions and found many BSD systems refuse installation. I believe the alphabet convention of a to z for a BSD-subpart makes BSD afraid of the integrity of the hard disk, it may not be able access it.

Therfore I think the problem of compatibility of Solaris/BSD system with Linux using extended partition is deep-rooted. It may be operationally one system cannot harm the other but accessing data across different platforms, thereby necessitating one system to read another system, may be a different ball game.

I use BSD systems, no apt-get here. It can bork pkg_add/pkg_delete (and /tmp noexec does bork build.sh, and many configure scripts, unless I properly set TMPDIR to point to a partition which allows exec)... but I am aware of these issues and remount as needed or set TMPDIR for building. If I am installing or deinstalling packages, I'll be remounting /usr anyway (to make it rw) so I just make /var exec for the duration of that task. And put everything back as it should be when done. But these tasks are not daily running and I just remember to do them as part of upgrades.

The idea is to make it more complicated (or rather make it less convenient) to change what is installed as part of the system -- for anyone so root gets included. So you'll most likely have to change a couple of these settings when doing administration and change them back when done. But for daily running of systems, these settings work well for me.

Don't confuse linux and its various partitioning tools limitations/peculiarities. Linux (kernel) has no problem with Solaris partitioning.
Correct, MS way of drive naming cannot give a clue about where the disk is. Linux is only partially better. hda6 doesn't give enough information to know where exactly this partition is located.
Grub is not linux. Grub is using a smart and non ambiguous way of naming the partitions, which is different from the one that Linux use internally.
Actually not. you can only have one primary partition that is subdivised this way. Unix is respecting (like MS) some old standard that requires a primary partition id to be only used at most once on a single hard disk. Linux doesn't care following this standard and allows multiple same IDs primary partitions.
I'm not sure about BSD, but Solaris is limiting by design the number of slices to 16.
As I already wrote, Linux isn't specifying in its partition numbering the kind of partition it is about (Unix slice or extended partition), however, this doesn't prevent it to be able to mount Solaris slices.

jlliagre,

It is nice to have someone to talk to about Solaris partition so I dwell on a bit if nobody minds.

I have a different opinion on this as I have installed Solaris 10, a BSD and a Solaris Express in the three primaries of the same hard disk before. If your statement is correct then there must be a mechanism within the Unix system to stop a user from installing the second Unix. I just haven't run into one yet.

I am interested in your asertion that Unix respecting the MS standard by using only one primary to dividing it into sub-slices as this proves Unix's utilisation is exactly as an extended partition. And as no two extended partition can exist in a PC then a Solaris must have (or to anticipate) conflicts with other PC operating systems installed into an extended partition.

To me if Solaris addresses a sub-slice as it allows Grub with a notation (hd1,4,a) then it implies the sub-slice can come from any of

Disk hd0, hd1, hd2....
Any of the 4 primary partition 1, 2, 3 or 4
Any of the sub-slice 1 to 16

This makes me think technically we can potentially put 4 different Solaris system in each hard dive with a mximum possible 64 sub-slices.

I always thought or speculated that this is the way Unix uses a hard disk and it is not going to compromise because the platform happens to be a PC. Tough if a MS system or a Linux can't cope with it but Unix can operate happily inside on its own.

Your last post makes me think again.

I also have no knowledge that a primary partition can have its own ID indicating it is a primary. AFAIK the 4 partitions listed in from the partition table located from 447 to 510th bytes of the MBR which is what the Bios reads and are commonly known as the primary partitions which every system uses in exactly the same way. I can't see what difference the way Linux usage is different from the others, except its naming convention is related to the hardware connections. For example hda is always the master disk of the primary IDE cable whereas hdc is the master of the secondary IDE acble even in a modern Bios any of the 4 IDE connection can be given the first bootable status.

It is not my intention to argue here but by comaparing our understanding of both the Linux and Solaris sides I may be able to find the gaps of my knowledge.

Just to clear things in my mind. Are there not 4 primary's possible to a disk? But, not so if extended partitions are created? correct? Sorry guys, the stuff about Solaris slices went over my head, but dont fret that, I don't run Solaris. Would love to find an older cheap? Sun workstation though.Just for interest sake.

As for Linux partitioning, I actually separate my system onto separate hard disks. 2.1Gb as /boot(100Mb) the rest swap. 20Gb /. 2x40Gb raided stripe ata /usr. 2x40Gb raided mirror ata /home/users.

I am not a sys op or an IT pro, just an avid hobbyist. Got sick of Winslows and its frailties long ago. Started with linux dual booting a 486\dx4 on W98SE and Caldera Non commercial. Really sad they went all SCO on us eh? That was the system that showed me what a real OS can do.

Just thought you might like some other ideas on partitioning schemes.
My 2 bob's worth, 2 cents worth, hell may even be worthless.

Shorty943.

It is the PC standard that there are 64 byes inside the MBR reserved for a partition table for 4 entries, 16 bytes each. Each entry has a record of its partition type (as type 6=Fat16, 83=native Linux, 82=swap, 7=ntfs etc), bootable flag indicating on or off, starting and finishing addresses in the hard disk and other essential information. These 4 entries are the primary partitions.

If a user wants more one of thses 4 entries must be given up to become an extended partition with the same information as a primary except the partition ID is different. In Linux a user can proceed to subdivides the entended partition into a maximum of 59 logical partition for a Pata disk but only 9 in the case of a Sata or SCSI disk.

The extended partition only has the starting point and finishing point and the Bios has no clue on how many logicals inside nor the size of any of them.

To find out the the make-up of the logical partitions the operating system must go to the first logical partition (from the starting point of the extended partition) position of the hard disk. Every logical in fact has its own reduced partition table inside which has a record of the address of the next logical partition in line. Thus logical partition must be continuous or the whole system breaks down.

In Linux you can see if one deletes a logical partition in the middle the space becomes dead and all the logical partitions after will shift up by one position without the user intervention.

It is possible that the way how a system finds a logical position makes it "rather difficult" if not impossible to have more than one extended partition. And this is why I am going to such a length to find out if Unix has a way of getting over such hurdle and can still be compatible with other PC system relying on the standard partitioning scheme.

There is IMO no issue with installing both Solaris and BSD on the same disks, as the partition IDs are different.
My statement wasn't about forbidding to have more than one subdivised partition of the same type.
Moreover, I doubt you were actually able to have both Solaris 10 and Solaris Express installed at the same time on the same disk. Unless you have changed one of partition IDs (and so disabled it). These is no way to access a slice outside of the first Solaris partition with the Solaris disk naming scheme (/dev/dsk/c0d0s*).
As I wrote, I was talking about same Unix, BSD and Solaris use different Ids, so can coexist.
Extended partitions are one partition type, BSD another and Solaris a third one, so they can coexist (I'm repeating myself ...).
This is true with the grub naming, but Solaris doesn't use it.
Neither do I.
That is correct.
Solaris and Linux do not use the same logic in naming the disks, nor do they provide the same level of details when naming extended partitions or slices. That was my point against the Linux way.

jlliagre ,

You are spot on in your prediction. I did hide the first Solaris when installing the second Solaris.

In booting Grub does the hiding of one and the unhiding of the other to boot whichever one I select. The hiding only alter the partition ID and the unhiding reverts its back. I have run Solaris and BSD with a bunch of Linux by hiding the entended partition and other unnecessary partitions to avoid conflict.

I use the same technique to run 3 Dos and 5 Windows in the same box too.

Thanks for confirming some of the Solaris characteristics for me, save a few blood stains on the brick wall.