| View previous topic :: View next topic |
| Author |
Message |
Rikishi42
Guest
|
 Posted: Tue Sep 23, 2008 3:12 am Post subject: Re: SCSI vs SATA hard disks |
 |
|
On 2008-09-22, Haines Brown <brownh@teufel.hartford-hwp.com> wrote:
| Quote: | I run my machine 24/7, and normally only reboot every few months to do
something with hardware. That should reduce the stress on the drive and
help it survive for a while.
|
Don't forget to backup. No matter how reliable the drives, no matter what
level of raid, backups are usefull.
--
Elevators smell different to midgets |
|
| |
|
 Back to top |
Aragorn
Guest
|
| Posted: Tue Sep 23, 2008 7:32 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Monday 22 September 2008 22:41, someone identifying as *Haines Brown*
wrote in /comp.os.linux.hardware:/
| Quote: | [...] I make my first partition at least 500 Mb so that I can do a base
installation in it using cdebootstrap. [...]
|
If you're concerned with having your system securely and properly set up,
then I would recommend that you split off as much from the (target) root
filesystem as possible at the root directory level and use custom mount
options for those filesystems.
Ideally, you should keep your root filesystem small - only about 250 MB max
- and static. Make sure */usr* is on a different filesystem and that it is
mounted read-only during normal system operation, and with /nodev,noatime/.
Likewise for */opt.*
Make the contents of */tmp* reside on /tmpfs/ instead of on a physical disk
and mount it with /noexec,nosuid,nodev./ Use a separate filesystem for
*/home* - it needn't be big on a server, although you may want to make it
larger if you intend using the machine as a workstation instead. Mount it
with /nodev,noatime./ */var* should also be separate and should be mounted
with /nodev,noatime./ */boot* can also be mounted read-only during normal
system operation - some even prefer not having it mounted at all - and
with /nodev,noatime/ as mount options.
Lastly, if you intend to use */srv* as a central storage for shared user
data or e.g. for the contents of a webserver running on that machine, make
it a separate filesystem as well, with /nodev,nosuid,noatime/ as mount
options.
You can determine the proper sizes for each of those filesystems using your
old hard disk as a template - give them about 10% slack just in case.
Keeping them all separate assures a more secure set up and will protect
your individual filesystems from filesystem corruption caused by some bug,
and from filesystem fragmentation on static filesystems such as the root
filesystem, */usr* and */opt.*
Finally this... If you need more than 15 partitions on the same disk (or
RAID array) and/or you would like to have resizeable filesystems, you
should look into logical volume management.
Your mileage may vary, but that's how I would do it. Actually, I *do* lay
out my systems like that. ;-)
--
*Aragorn*
(registered GNU/Linux user #223157) |
|
| |
|
| Back to top |
david
Guest
|
| Posted: Tue Sep 23, 2008 9:25 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Tue, 23 Sep 2008 11:00:59 +0200, Pascal Hambourg rearranged some
electrons to say:
| Quote: | Aragorn a écrit :
but either way, even 300 MB/sec by far exceeds the real throughput
capacity of a single hard disk, and with both SAS and SATA, the
available bus speed applies to each individual storage device,
Well, to my dispair I read that SATA III allows multiple devices on a
single bus. Doh.
|
There is no such thing as SATA III. There is SATA 6GB, and rev 3.0 of
the specification.
http://www.serialata.org/6gbnamingguidelines.asp
Where did you read that the next revision of the SATA spec will allow
multiple devices per bus? because that is not per the SATA specification,
which is a point to point connection.
See the official SATA web site:
http://www.sata-io.org/satatechnology.asp
| Quote: | as opposed
to for the total of devices connected to the controller as in the case
with parallel SCSI or ATA. ;-)
s/controller/channel or bus/ |
|
|
| |
|
| Back to top |
Pascal Hambourg
Guest
|
| Posted: Tue Sep 23, 2008 2:00 pm Post subject: Re: SCSI vs SATA hard disks |
|
|
Aragorn a écrit :
| Quote: |
but either way, even 300 MB/sec by far exceeds the real
throughput capacity of a single hard disk, and with both SAS and SATA, the
available bus speed applies to each individual storage device,
|
Well, to my dispair I read that SATA III allows multiple devices on a
single bus. Doh.
| Quote: | as opposed
to for the total of devices connected to the controller as in the case with
parallel SCSI or ATA. 
|
s/controller/channel or bus/ |
|
| |
|
| Back to top |
Pascal Hambourg
Guest
|
| Posted: Tue Sep 23, 2008 3:46 pm Post subject: Re: SCSI vs SATA hard disks |
|
|
david a écrit :
| Quote: | On Tue, 23 Sep 2008 11:00:59 +0200, Pascal Hambourg rearranged some
electrons to say:
Well, to my dispair I read that SATA III allows multiple devices on a
single bus. Doh.
There is no such thing as SATA III. There is SATA 6GB, and rev 3.0 of
the specification.
|
Right, SATA III is a misnomer.
| Quote: | Where did you read that the next revision of the SATA spec will allow
multiple devices per bus? because that is not per the SATA specification,
which is a point to point connection.
|
In the Wikipedia french article about Serial ATA.
<http://fr.wikipedia.org/wiki/Serial_ATA#Version_III_du_Serial_ATA>
I'll be glad if it is wrong. The english article only mentions port
expanders as a means to connect multiple devices to a single controller
port. This preserves the point to point nature of the SATA bus, but may
cause bus contention when multiple devices are simulaneously active. |
|
| |
|
| Back to top |
Sheridan Hutchinson
Guest
|
| Posted: Tue Sep 23, 2008 4:20 pm Post subject: Re: SCSI vs SATA hard disks |
|
|
Aragorn wrote:
| Quote: | Finally this... If you need more than 15 partitions on the same disk
(or RAID array) and/or you would like to have resizeable filesystems,
you should look into logical volume management.
|
I'm always looking to learn something new, so I wonder if you could
share with me your experience of the benefits from having multiple
partitions, as opposed to having a single partition housing all the
mount points?
Then reason I ask is that I have a desktop and laptop with encrypted
LVM's that house / and a swap in separate logical volume, but within the
same logical group. The only other linux partitions is /boot which of
course needs to remain unencrypted.
I backup regularly using full disks clones for easy restoration, so I'm
never worried about losing data.
So just in theory, lets pretend I didn't have an encrypted LVM, what
would be the benefits be?
--
Regards,
Sheridan Hutchinson
Sheridan@Shezza.org |
|
| |
|
| Back to top |
Haines Brown
Guest
|
| Posted: Tue Sep 23, 2008 5:03 pm Post subject: Re: SCSI vs SATA hard disks |
|
|
Aragorn <aragorn@chatfactory.invalid> writes:
| Quote: | On Monday 22 September 2008 22:41, someone identifying as *Haines Brown*
wrote in /comp.os.linux.hardware:/
[...] I make my first partition at least 500 Mb so that I can do a base
installation in it using cdebootstrap. [...]
If you're concerned with having your system securely and properly set up,
then I would recommend that you split off as much from the (target) root
filesystem as possible at the root directory level and use custom mount
options for those filesystems.
|
No, I actually have 12 partitions at present that are broken out (3 of
which are custom). The root was made large so that it can serve as a
temporary installation partition, from which directories will be moved
out once the base system is in place there. However, it makes more sense
to use the swap partition for this purpose because these days it is
plenty big. I run a workstation rather than a server.
You bring up some interesting points about partitioning policy. Although
OT, I wonder how to implement /tmp as a tmpfs rather than ext3
on the physical disk and why this it is a good idea to do it.
An issue that has always troubled me is the optimal sequence of
partitions. For example, /swap is supposed to be located in relation to
the physical disks on the hard disk, but I never knew how to do
that. Howver, I'm not sure it makes any difference any more.
--
Haines Brown, KB1GRM |
|
| |
|
| Back to top |
Hactar
Guest
|
| Posted: Tue Sep 23, 2008 9:37 pm Post subject: Re: SCSI vs SATA hard disks |
|
|
In article <87zllz9iip.fsf@teufel.hartford-hwp.com>,
Haines Brown <brownh@teufel.hartford-hwp.com> wrote:
| Quote: | Aragorn <aragorn@chatfactory.invalid> writes:
On Monday 22 September 2008 22:41, someone identifying as *Haines Brown*
wrote in /comp.os.linux.hardware:/
[...] I make my first partition at least 500 Mb so that I can do a base
installation in it using cdebootstrap. [...]
If you're concerned with having your system securely and properly set up,
then I would recommend that you split off as much from the (target) root
filesystem as possible at the root directory level and use custom mount
options for those filesystems.
No, I actually have 12 partitions at present that are broken out (3 of
which are custom). The root was made large so that it can serve as a
temporary installation partition, from which directories will be moved
out once the base system is in place there. However, it makes more sense
to use the swap partition for this purpose because these days it is
plenty big. I run a workstation rather than a server.
You bring up some interesting points about partitioning policy. Although
OT, I wonder how to implement /tmp as a tmpfs rather than ext3
on the physical disk
|
For example, "mount none /mnt/temp -t tmpfs -o size=10M"
| Quote: | and why this it is a good idea to do it.
|
That's tougher. Frees up disk space? No, disk is much cheaper than
RAM. Speeds up some apps? If you reboot weekly at least you can scrap
tmpclean or whatever it's called.
| Quote: | An issue that has always troubled me is the optimal sequence of
partitions. For example, /swap is supposed to be located in relation to
the physical disks on the hard disk, but I never knew how to do
that. Howver, I'm not sure it makes any difference any more.
|
Why wouldn't it make any difference? What I wonder is: are disk
cylinder numbers arranged like
1 - 511
---------+---------
512 -1023
1024-1535
---------+---------
1536-2048
or like
1 - 511
---------+---------
1023- 512
1024-1535
---------+---------
2048-1536
or like
1 2045
--------+---------
2 2046
3 2047
--------+---------
4 2048
or does it vary by disk line? Heck, when the pool of bad block
replacements is used, you're almost guaranteed an extra seek.
--
-eben QebWenE01R@vTerYizUonI.nOetP royalty.mine.nu:81
VIRGO: All Virgos are extremely friendly and intelligent - except
for you. Expect a big surprise today when you wind up with your
head impaled upon a stick. -- Weird Al, _Your Horoscope for Today_ |
|
| |
|
| Back to top |
Aragorn
Guest
|
| Posted: Wed Sep 24, 2008 6:04 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Tuesday 23 September 2008 14:03, someone identifying as *Haines Brown*
wrote in /comp.os.linux.hardware:/
| Quote: | Aragorn <aragorn@chatfactory.invalid> writes:
On Monday 22 September 2008 22:41, someone identifying as *Haines Brown*
wrote in /comp.os.linux.hardware:/
[...] I make my first partition at least 500 Mb so that I can do a base
installation in it using cdebootstrap. [...]
If you're concerned with having your system securely and properly set up,
then I would recommend that you split off as much from the (target) root
filesystem as possible at the root directory level and use custom mount
options for those filesystems.
No, I actually have 12 partitions at present that are broken out (3 of
which are custom). The root was made large so that it can serve as a
temporary installation partition, from which directories will be moved
out once the base system is in place there. However, it makes more sense
to use the swap partition for this purpose because these days it is
plenty big. I run a workstation rather than a server.
|
Or alternatively, you can create a second root partition, which you leave
unmounted once the system is set up properly. That way, you have a rescue
root filesystem for in the event something goes wrong with your main root
filesystem.
| Quote: | You bring up some interesting points about partitioning policy. Although
OT, I wonder how to implement /tmp as a tmpfs rather than ext3
on the physical disk and why this it is a good idea to do it.
|
Doing it is rather easy. Simply mount /tmpfs/ to it. Boot up in single
user mode, make sure */tmp* is empty and then and edit your */etc/fstab* to
include a line like this:
none /tmp tmpfs auto,nouser,nodev,noexec,nosuid,noatime 0 0
If so desired, you can also easily control the maximum allowed size of
*/tmp* in this manner by adding the /size=/ mount option - see the /man/
pages for details - so as to limit the available space for */tmp* without
having to set up quota or resize on-disk filesystems. The default maximum
size if no parameter is given will be half your available RAM.
The reason as to why one should make */tmp* a /tmpfs* is that according to
the FHS 2.3, nothing in */tmp* should be expected to survive a reboot.
It's normally only needed for sockets and such, and those things don't take
up much diskspace.
| Quote: | An issue that has always troubled me is the optimal sequence of
partitions. For example, /swap is supposed to be located in relation to
the physical disks on the hard disk, but I never knew how to do
that. Howver, I'm not sure it makes any difference any more.
|
Normally, manuals and installers will encourage you to create a swap
partition right behind the root filesystem - so as to make sure that you
don't forget creating one - but on systems with a lot of RAM, the location
on disk of the swap partition is not really relevant anymore.
It used to be relevant on older systems with less RAM because the closer the
swap partition is to the start of the hard disk, the faster it will be, due
to the fact that the outermost cylinders have more sectors on them than the
innermost ones, and thus more sectors can be read in one go without having
to switch heads or move the position of the heads over the platters.
I normally create */boot* as the first partition, followed by the root
filesystem itself. And as a third, I will then create either */usr* or the
swap partition. It has always proven to be a good set-up. :-)
--
*Aragorn*
(registered GNU/Linux user #223157) |
|
| |
|
| Back to top |
Aragorn
Guest
|
| Posted: Wed Sep 24, 2008 6:29 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Tuesday 23 September 2008 13:20, someone identifying as *Sheridan
Hutchinson* wrote in /comp.os.linux.hardware:/
| Quote: | Aragorn wrote:
Finally this... If you need more than 15 partitions on the same disk
(or RAID array) and/or you would like to have resizeable filesystems,
you should look into logical volume management.
I'm always looking to learn something new, so I wonder if you could
share with me your experience of the benefits from having multiple
partitions, as opposed to having a single partition housing all the
mount points?
|
Well, as I mentioned in my previous post, there are many benefits...:
(1) You can keep your static trees static. No fragmentation caused by
mixing variable files with static files.
(2) You reduce the risk of total filesystem corruption. If something
does go wrong and you end up with filesystem corruption in a given
partition, then this need not necessarily be the case for the other
filesystems on your computer.
(3) In the event of reinstallation of the operating system, you can keep
your data separate, and the filesystems holding them - e.g. */home* -
need not be reformatted.
(4) Security. You can have your static filesystems mounted read-only
during normal system operation, and so not even a rogue root process
would be able to write to your system directories, because they'd
have to be remounted in read/write mode by the root user first. In
addition, this also plays nicely with (2) higher up, i.e. a filesystem
mounted read-only cannot get corrupted by - say - an unclean shutdown
due to an unexpected power failure.
(5) Related to (4), you can increase security by using separate mount
options for each filesystem, and you can even use different filesystem
types for individual needs, if you you choose to. For instance, on a
server that offers streaming audio or video, you will most likely use
an /xfs/ or /jfs/ filesystem for the data that is to be streamed. By
using different partitions, you can then also specify specific block
sizes while formatting the filesystems, depending on what you'll be
using them for.
The downside is that the above incorporates an additional level of
complexity that may seem daunting to the newbie - read: the Windows
habituate - because this kind of stuff is simply unheard of in Windows.
Therefore, desktop-oriented GNU/Linux distributions typically default to
using just a root filesystem and a swap partition, eventually with a
separate */home.* The advantage of this simplified approach is also that
the newbie need not concern himself/herself with partition sizes.
On the other hand, one can use a distributed filesystem hierarchy in which
partition sizes are variable, by creating a fixed partition for */boot*,
the root filesystem and the swap partition, and then creating an extra
partition in which individual filesystems are created in resizable logical
volumes.
The caveat in the last above suggestion however is that you cannot put
*/boot* on a logical volume unless you're using LILO as your boot loader,
installed in the MBR, due to the fact that the GRUB bootloader needs a
filesystem driver in order to load the kernel and is unfamiliar with
logical volumes. For LILO, this is not true because LILO uses a hardcoded
logical block address to reach the on-disk location of the kernel.
Additionally, having the root filesystem live on a logical volume, you'll
have to boot with an /initrd/ with LVM support built-in, but of course most
GNU/Linux distributions come with highly modular stock kernels that require
an /initrd/ anyway. It does however serve to be noticed for in the event
that you roll your own kernels. ;-)
| Quote: | Then reason I ask is that I have a desktop and laptop with encrypted
LVM's that house / and a swap in separate logical volume, but within the
same logical group. The only other linux partitions is /boot which of
course needs to remain unencrypted.
|
With regard to volume groups, I recommend using one and the same volume
group for all your logical volumes on the same hard disk, unless you have a
good reason as to why you would need more volume groups.
I myself have such a reason on my other machine, namely that I use different
volume groups to group together the filesystems belonging to the same
virtual machine, as that physical computer is going to run four virtual
machines with Xen, and each of those virtual machines will have highly
distributed filesystem hierarchies. With such a set-up, it's easier to
remove all filesystems belonging to a single virtual machine by simply
removing the volume group. ;-)
| Quote: | I backup regularly using full disks clones for easy restoration, so I'm
never worried about losing data.
|
Backups are a necessity, but that still doesn't mean that you can't do more
to protect your data and your system integrity. ;-)
| Quote: | So just in theory, lets pretend I didn't have an encrypted LVM, what
would be the benefits be?
|
I believe I've covered that higher up in this reply already. ;-)
--
*Aragorn*
(registered GNU/Linux user #223157) |
|
| |
|
| Back to top |
Aragorn
Guest
|
| Posted: Wed Sep 24, 2008 6:33 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Tuesday 23 September 2008 18:37, someone identifying as *Hactar* wrote
in /comp.os.linux.hardware:/
| Quote: | [...] What I wonder is: are disk cylinder numbers arranged like
1 - 511
---------+---------
512 -1023
1024-1535
---------+---------
1536-2048
or like
1 - 511
---------+---------
1023- 512
1024-1535
---------+---------
2048-1536
or like
1 2045
--------+---------
2 2046
3 2047
--------+---------
4 2048
or does it vary by disk line? Heck, when the pool of bad block
replacements is used, you're almost guaranteed an extra seek.
|
True, and then you're not even keeping into account that the disk geometry
as reported to the kernel is not the actual geometry at all, because with
modern-day large hard disks it is always being translated. And then
there's the low-level format, and such.
Modern hard disks are to be seen as abstractions, just like when you have a
RAID array that is seen by the kernel as being a single disk. The
optimizing is all already done in hardware.
--
*Aragorn*
(registered GNU/Linux user #223157) |
|
| |
|
| Back to top |
Sheridan Hutchinson
Guest
|
| Posted: Wed Sep 24, 2008 6:41 am Post subject: Re: SCSI vs SATA hard disks |
|
|
Aragorn wrote:
<snip>
| Quote: | I believe I've covered that higher up in this reply already.
|
And indeed you did! Thank you for such a well-considered reply, I shall
print it and consider the ramifications of making some changes to my
machines :)
--
Regards,
Sheridan Hutchinson
sheridan@shezza.org |
|
| |
|
| Back to top |
Haines Brown
Guest
|
| Posted: Wed Sep 24, 2008 7:08 am Post subject: Re: SCSI vs SATA hard disks |
|
|
ebenZEROONE@verizon.net (Hactar) writes:
| Quote: | In article <87zllz9iip.fsf@teufel.hartford-hwp.com>,
Haines Brown <brownh@teufel.hartford-hwp.com> wrote:
An issue that has always troubled me is the optimal sequence of
partitions. For example, /swap is supposed to be located in relation to
the physical disks on the hard disk, but I never knew how to do
that. Howver, I'm not sure it makes any difference any more.
Why wouldn't it make any difference? What I wonder is: are disk
cylinder numbers arranged like
|
Yes, but how does one know what the physical platter limits are so that
one can put a partion near the edge?
Also, these days, I assume one does not normally use swap space, and so
what effect would its location normally have? What partitions need to be
near the platter edges?
--
Haines Brown, KB1GRM |
|
| |
|
| Back to top |
Aragorn
Guest
|
| Posted: Wed Sep 24, 2008 7:08 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Wednesday 24 September 2008 04:33, someone identifying as *Haines Brown*
wrote in /comp.os.linux.hardware:/
| Quote: | ebenZEROONE@verizon.net (Hactar) writes:
In article <87zllz9iip.fsf@teufel.hartford-hwp.com>,
Haines Brown <brownh@teufel.hartford-hwp.com> wrote:
An issue that has always troubled me is the optimal sequence of
partitions. For example, /swap is supposed to be located in relation to
the physical disks on the hard disk, but I never knew how to do
that. Howver, I'm not sure it makes any difference any more.
Why wouldn't it make any difference? What I wonder is: are disk
cylinder numbers arranged like
Yes, but how does one know what the physical platter limits are so that
one can put a partion near the edge?
|
With a tool like /cfdisk/ you can create a partition that's aligned with the
end of the available diskspace, and thus close to the center of the platter
surface.
There are typically a few more cylinders beyond that, but those are not
coated with a magnetically susceptible alloy and are used to allow the disk
heads to land. (Hitachi/IBM disks are an exception to this rule as they
park the diskheads on a ramp outside the platter circumference.)
| Quote: | Also, these days, I assume one does not normally use swap space, and so
what effect would its location normally have? What partitions need to be
near the platter edges?
|
The ones that require the least I/O.
--
*Aragorn*
(registered GNU/Linux user #223157) |
|
| |
|
| Back to top |
Aragorn
Guest
|
| Posted: Wed Sep 24, 2008 7:08 am Post subject: Re: SCSI vs SATA hard disks |
|
|
On Wednesday 24 September 2008 03:41, someone identifying as *Sheridan
Hutchinson* wrote in /comp.os.linux.hardware:/
| Quote: | Aragorn wrote:
I believe I've covered that higher up in this reply already. ;-)
And indeed you did! Thank you for such a well-considered reply, I shall
print it and consider the ramifications of making some changes to my
machines 
|
Glad I could help! ;-)
--
*Aragorn*
(registered GNU/Linux user #223157) |
|
| |
|
| Back to top |
|
