PDA

View Full Version : Re: Quick question for NutCracker...


NuT CrAcKeR
May 1st 04, 04:39 PM
To the best of my knowledge, there is no USB hearders on that server.

There is an easy way to check without a risk to your uptime: look at the
back of the server.

You should be able to make due with removing the USB support from the kernel
and recompiling.

Yes, you should have a 3200 or better, and also there is nothing preventing
you from ADDING a usb card to the server if you wish.

Cheers,

LC

"MattD.." > wrote in message
...
> ...or anyone else in the know.
>
> I have a Compaq Proliant 6500 Xeon (Dual PIII Xeon 500, 1.2GB memory,
Smart
> 2/DH RAID (yes, I know it should have a 3200 - If anyone wants to donate
> one, or better still a 4200 64 bit, I'll put it right!), new style 9 drive
> bay with 6 populated, Netelligent TI Thunderlan NIC etc.) running FreeBSD
> 5.2.1.
>
> Please ignore the BSD bit. This question is hardware only.
>
> In the dmesg, I get this:
> uhci0: <Intel 82371AB/EB (PIIX4) USB controller> port 0x2c20-0x2c3f at
> device 15.2 on pci0
> uhci0: Could not allocate irq
>
> Now, I know why it doesn't allocate an IRQ. It hasn't been detected by the
> <F10> setup utility and its PnP details configured and stored in NVRAM.
The
> question is, does this controller have a physical presence on the board
> (molex header etc.) or is it just an unused feature of the chipset Compaq
> have used? It would be handy to have a USB port for a couple of reasons,
> specifically UPS interfacing, so if it can be brought out to the rear
> panel, I'd like to do it.
>
> Also, will the latest setup utility recognize the controller? I used the
4.7
> Smart Start to set up both the initial config and the system partition,
> which is getting on a bit now. OK, I could try it, but I'd like to know if
> it's available as a connection first before risking my month+ uptime
figure
> (I've only had the thing 6 weeks, so reliability is looking good) on a
> whim. If it isn't available, I'll take the uhci device out of the kernel.
I
> don't like dmesg errors, especially when trying to convince clients that
> BSD is the way to go ;o)
>
> Any advice gratefully received. Everything else works perfectly and having
> used nothing but Compaq/BSD combinations as servers for a good while, I'm
> pretty au-fait with the nuances.
> --
> MattD..
> mattd145 <at> onetel <dot> com

NuT CrAcKeR
May 1st 04, 09:21 PM
"MattD.." > wrote in message
...
> On Saturday 01 May 2004 16:39, the murky waters churned and seethed, the
> dark weeds parted and the water took on the sinister, shifting visage of
> NuT CrAcKeR. The great maw opened and the following was heard:
>
> > To the best of my knowledge, there is no USB hearders on that server.
>
> Not even internally?

no, not even internally. HPaq isnt really big on providing internal headers
to things. You might be thinking of motherboard architectures you would find
in a PC... To give you some perspective, even the DL360 G1 has a space on
the system board reserved for USB, but no header. Its a shame really,
because the pizza-boxes are ideal for them.

> > There is an easy way to check without a risk to your uptime: look at the
> > back of the server.
>
> That's what prompted the question. No USB ports on the back, but the dmesg
> lines indicating the presence of an Intel USB device. I'll assume it's an
> unimplemented chipset feature...
>
> > You should be able to make due with removing the USB support from the
> > kernel and recompiling.
>
> ...and do just this.
>
> > Yes, you should have a 3200 or better, and also there is nothing
> > preventing you from ADDING a usb card to the server if you wish.
>
> Yep, that was my thinking, although with five hot-plug PCI slots to play
> with, I would rather have had USB on the board and a slot free for other
> things. I will get around to a better RAID controller (I think I'll just
> get the 4200 anyway. There seem to be enough of them around) one of these
> days, at which point I'll bit the bullet, recompile my kernel without
> device uhci and replace the controller all at once. For now I'll just pipe
> dmesg through grep in my .cshrc so the uhci0 lines don't appear in front
of
> the beancounters ;o)

dont let the hotplug slots lead you astray. all it means is that if a card
that is already installed fails, you can replace it with an identical card
without downing the server. Its a nice feature for sure, but the only place
i would consider actually using a hotplug card would be for an HBA to a SAN
(in a mulitpath configuration) or for redunant NIC's or (cringe) Array
controllers.

Unless you have driped some serious cash on the rest of your infrastructure
and hardware, hotplug slots are kinda a fallacy.

>
> Thanks for the advice.

You are welcome, glad I could help.

> --
> MattD..
> mattd145 <at> onetel <dot> com

NuT CrAcKeR
May 2nd 04, 02:47 AM
Matt,

Hotplug PCI has been standard equipment on enterprise servers for at least 7
years and your distro hasnt implimented supprot for it yet? You may want to
consider running one of the *nix's that are supported on the compaq's. They
have hotplug support built into the drivers for thier devices, so you can
have the same robust experience with thier machines no matter what
(supported) OS you are running.

Its also awsome to have all the insight bells and whistels available to you.
If you arent usging them, you are cheating yourself.

LC

"MattD.." > wrote in message
...
> On Saturday 01 May 2004 21:21, the murky waters churned and seethed, the
> dark weeds parted and the water took on the sinister, shifting visage of
> NuT CrAcKeR. The great maw opened and the following was heard:
>
> > dont let the hotplug slots lead you astray. all it means is that if a
card
> > that is already installed fails, you can replace it with an identical
card
> > without downing the server. Its a nice feature for sure, but the only
> > place i would consider actually using a hotplug card would be for an HBA
> > to a SAN (in a mulitpath configuration) or for redunant NIC's or
(cringe)
> > Array controllers.
> >
> > Unless you have driped some serious cash on the rest of your
> > infrastructure and hardware, hotplug slots are kinda a fallacy.
>
> Yep, but it's a non-op anyway since BSD doesn't yet support hot-plug PCI.
> There's a few messages flying about on the -CURRENT mailing list about
> future support for hot-plug PCI, but it's a way off yet, probably for
> exactly the resons you state. It's just not essential unless you really
> *must* have 0% downtime. BSD boots fast enough (once you find out how to
> stop it waiting 15 seconds for SCSI devices to settle and bring that down
> to 3-5 seconds just in case) to allow a quick change of anything that dies
> anyway. The only reason I try not to reboot if I can help it is the server
> is a YP/NIS master and local split-brain DNS.
>
> Thanks again for your help. It saved me a lot of time searching HP's site
> and taking the Proliant to bits.
>
> --
> MattD..
> mattd145 <at> onetel <dot> com