What *is* this silliness? Yes, the out of box configurations are smaller,
but most certainly configure things larger- Unix has always been this broken-
even AIX, although at least with AIX you could use SMIT to pump things up w/o
rebooting.
You can't fit more than 126 devices on a local loop. So what? That's a
limitation of FC-AL, not Linux. A more important limit for the Qlogic cards is
the inability to maintain fabric logins for more than about ~124 devices at a
time, so you have essential a tlb type of problem and no midlayer support that
gives you sensible policy as to which are the interesting devices at any given
point.
In terms of lun support- this is, as of 2.2.14 for sure- a property of the
HBA merged with a partially kernel configurable and certainly patchable and
eddittable value in scsi.c (max_scsi_luns). In fact, the max_luns property in
the host adapter structure is an unsigned int. There is a problem that the
arguments to scan_scsis are unsigned char, which puts Linux on par with the
non-NDI framework Solaris, but this is so trivial a fix that it's hardly worth
discussing. In terms of adding extra devs- this is a quantity only pertinent
for allocating extra space for loadable HBA rescans- this doesn't have
anything to do with boot time probing with static linking, or if it does, you
can always change the SD_EXTRA_DEVS to something reasonable (like 512).
But this thread started out with comparisons to Solaris, and frankly, Solaris
(out of the box) isn't in any better shape than Linux is (which I pointed out)
with respect to FC midlayer stuff, or storage management (with the default end
user product). It's only when you add some adult commercial products (like
Veritas stuff) that this stuff becomes managable.
Along the same vein, Linux becomes more interesting when you add Heinz' LVM
stuff, or possibly Richard Gooch's devfs. The former is more production
quality as far as I know- I don't know too much yet about the latter. So in
terms of full FC midlayer management, who, other than possibly 5.0 Tru64
(which was supposed to have some of this) has it? In terms of large device
pool management, Linux is in no worse shape as an out of the box product than
Solaris w/o some of these newer addons (but less than, say, FreeBSD, or HP/UX
or IRIX or AIX).
-matt
On Wed, 8 Mar 2000 [EMAIL PROTECTED] wrote:
>
>
>
>
> >There *IS* one limitation in the Linux SCSI subsystem that may be biting
> >you -- it only addresses up to 16 devices. That's all, folks! Obviously
> >this is something that has to be changed before Linux can be taken
> >seriously in the high-end storage department.
>
> There are lots of caveats to the above mentioned 16 device limit. RH6.1
> (kernel 2.2.12-20) will do more than 16 devices/LUNs IF you run the
> /dev/MAKEDEV script to create more devices ("out of box" stops at
> /dev/sdpX).
>
> The real killer is the limit of 4 devices on a secondary SCSI bus due to
> SD_EXTRA_DEVS. I'm testing a 11.2TB subsystem for Linux compatibility.
> You ever tried to hide 11.2TB behind a grand total of FOUR LUNS?!? ;)
> I've heard future releases will take this to 40, but you can fit
> 127 devices on a FC-AL. sigh.
>
> >However, a RAID box *SHOULD* present its dozens of devices as a single
> >device to the Linux kernel... so I don't know why this would be happening.
>
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
