Autoreply: Beowulf digest, Vol 1 #570 - 7 msgs

Sat Sep 8 13:04:01 PDT 2001

The user that you sent email to does not exist on this server. Please check the email address and try again.

Your message reads:

Date: Sat, 8 Sep 2001 12:00:07 -0400
Message-Id: <200109081600.MAA19814 at blueraja.scyld.com>
From: beowulf-admin at beowulf.org
Subject: Beowulf digest, Vol 1 #570 - 7 msgs
Reply-to: beowulf at beowulf.org
Mime-version: 1.0
Content-type: text/plain
To: beowulf at beowulf.org
Sender: beowulf-admin at beowulf.org
Errors-To: beowulf-admin at beowulf.org
Precedence: bulk

Send Beowulf mailing list submissions to
	beowulf at beowulf.org

To subscribe or unsubscribe via the web, visit
	http://www.beowulf.org/mailman/listinfo/beowulf
or, via email, send a message with subject or body 'help' to
	beowulf-request at beowulf.org
You can reach the person managing the list at
	beowulf-admin at beowulf.org

When replying, please edit your Subject line so it is more specific than
"Re: Contents of Beowulf digest..."

Today's Topics:

  1. Re:WOL: how does it work? (Greg Lindahl)
  2. Re:WOL: how does it work? (Claude Pignol)
  3. Re:WOL: how does it work? (Robert G. Brown)
  4. Re:WOL: how does it work? (Donald Becker)
  5. Re:WOL: how does it work? (Donald Becker)
  6. Price of a 24 cluster (Luismar Marques Porto)
  7. RE:D-Link switch for b at h (Martin WHEELER)

--__--__--

Message: 1
Date: Fri, 7 Sep 2001 16:14:49 -0400
From: Greg Lindahl <lindahl at conservativecomputer.com>
To: beowulf at beowulf.org
Subject: Re: WOL: how does it work?

On Fri, Sep 07, 2001 at 12:20:48PM -0700, Martin Siegert wrote:

> Here is the problem: Each node draws a current of about 1.5A (I measured
> that a few days ago). Since I have about 70 of those, booting all nodes
> at once will draw all of the sudden a current of more than 100A. The
> people who run our machine room don't allow me to do that (probably for
> good reason).

I can't really answer your question, but there's an alternate
solution. You can use a device which delays the booting of some
nodes. For example, the APC MasterSwitch has the ability to let you
power cycle nodes by attaching to a web browser, but another feature
is that it can power up the nodes with a delay after a power failure.
It's a bit expensive for this purpose ($354 for 8 plugs @ 120V, 12A
total), but maybe you can find something cheaper, such as an X10 based
controller.

greg

--__--__--

Message: 2
Date: Fri, 07 Sep 2001 16:23:27 -0500
From: Claude Pignol <cpignol at seismiccity.com>
Organization: SeismicCity, Inc.
To: Martin Siegert <siegert at sfu.ca>
CC: beowulf <beowulf at beowulf.org>
Subject: Re: WOL: how does it work?
boundary="------------060806040101010806080706"

--------------060806040101010806080706
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

I think the wake up depends on how the node enter the  sleeping state.
May be you could try this with the original kernel of Beowulf 27bz-7
power down remotely a node:
bpctl -S Node_Number -s pwroff

The node shutdown and the power off but the NIC is alive
Don't remove the power plug. (Important)

send the magic packet

ether-wake -i eth4 00:E0:81:03:21:DD

It should wake up the node
If not add the kernel parameter
apm=power-off
to the slave node kernel

And restart the whole process.

This works fine for me (but It's not the same motherboard)
I hope this could help
Claude

Martin Siegert wrote:

>I am trying to get wake-on-lan (WOL) to wirk on my Beowulf cluster
>and until now I failed. Admittedly I don't know much about WOL, thus
>this failure may just be due to some stupid mistake on my part.
>
>Here is the problem: Each node draws a current of about 1.5A (I measured
>that a few days ago). Since I have about 70 of those, booting all nodes
>at once will draw all of the sudden a current of more than 100A. The
>people who run our machine room don't allow me to do that (probably for
>good reason). Thus I decided on the following approach:
>
>The bios for the motherboard that I'm using (Tyan Thunder K7) allows
>two setting for what to do after a power failure when the power comes
>back on: a) stay off or b) power on.
>Instead of choosing b) for all nodes (wich would cause the aforementioned
>problem)  I want to choose b) only for the master node and a) for all slaves.
>Then use WOL from the master node to wake up the slave sequentially
>using a script and the ether-wake program from
>http://www.scyld.com/expert/wake-on-lan.html.
>
>Unfortunately, I have been unable to wake up a node. Here is what I do:
>"halt" a node. Detach the power cable. Reattach the power cable.
>At this point the lights on the two onboard NICs (the Tyan web site
>and the printing on the chips say that those are 3c920, the 3c59x driver
>identifies them as 3c980; I don't know whether that is relevant; the NICs
>work fine) come on. A Tyan technician told me that WOL on the Thunder K7 is
>always on, no special BIOS setup would be needed. They also told me that I
>have to use a 2.4.x kernel because only those would support APCI. I don't
>understand why the kernel is important here: when the node is halted
>what difference does the kernel make for the receiving of the magic WOL
>packet that is supposed to wake up the box? Anyway, I compiled a
>2.4.9-ac8 kernel with APIC enabled, which I use with the "noapic"
>kernel option in /etc/lilo.conf. I have also tried the stock RH 7.1
>2.4.3-12smp kernel without any difference with respect to WOL (i.e.,
>no success).
>
>After reattaching the power cable I then send the magic packet from
>the master node:
>
>./ether-wake -i eth4 00:E0:81:03:21:DD
>
>where 00:E0:81:03:21:DD is the MAC address of one of the onboard NICs
>on the node. tcpdump shows that the packet actually is sent. Also the
>lights in the NICs on the sending and receiving end flash, but otherwise
>nothing happens.
>
>What's wrong? Any suggestions are most appreciated.
>
>Thanks!
>
>Martin
>
>========================================================================
>Martin Siegert
>Academic Computing Services                        phone: (604) 291-4691
>Simon Fraser University                            fax:   (604) 291-4242
>Burnaby, British Columbia                          email: siegert at sfu.ca
>Canada  V5A 1S6
>========================================================================
>
>_______________________________________________
>Beowulf mailing list, Beowulf at beowulf.org
>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>

-- 
------------------------------------------------------------------------
Claude Pignol SeismicCity, Inc. <http://www.seismiccity.com>
2900 Wilcrest Dr.    Suite 470  Houston TX 77042
Phone:832 251 1471 Mob:281 703 2933  Fax:832 251 0586

--------------060806040101010806080706
Content-Type: text/html; charset=us-ascii
Content-Transfer-Encoding: 7bit

<html>
<head>
</head>
<body>
I think the wake up depends on how the node enter the  sleeping state.<br>
May be you could try this with the original kernel of Beowulf 27bz-7<br>
power down remotely a node:<br>
bpctl -S Node_Number -s pwroff<br>
<br>
The node shutdown and the power off but the NIC is alive<br>
Don't remove the power plug. (Important)<br>
<br>
send the magic packet
<pre wrap="">ether-wake -i eth4 00:E0:81:03:21:DD</pre>
It should wake up the node<br>
If not add the kernel parameter <br>
apm=power-off<br>
to the slave node kernel<br>
<br>
And restart the whole process.<br>
<br>
This works fine for me (but It's not the same motherboard)<br>
I hope this could help<br>
Claude<br>
<br>
<br>
  <br>
<br>
Martin Siegert wrote:<br>
<blockquote type="cite" cite="mid:20010907122048.A31459 at stikine.ucs.sfu.ca">
  <pre wrap="">I am trying to get wake-on-lan (WOL) to wirk on my Beowulf cluster<br>and until now I failed. Admittedly I don't know much about WOL, thus<br>this failure may just be due to some stupid mistake on my part.<br><br>Here is the problem: Each node draws a current of about 1.5A (I measured<br>that a few days ago). Since I have about 70 of those, booting all nodes<br>at once will draw all of the sudden a current of more than 100A. The<br>people who run our machine room don't allow me to do that (probably for<br>good reason). Thus I decided on the following approach:<br><br>The bios for the motherboard that I'm using (Tyan Thunder K7) allows<br>two setting for what to do after a power failure when the power comes<br>back on: a) stay off or b) power on.<br>Instead of choosing b) for all nodes (wich would cause the aforementioned<br>problem)  I want to choose b) only for the master node and a) for all slaves.<br>Then use WOL from the master node to wake up the slave seq
uentially<br>using a script and the ether-wake program from<br><a class="moz-txt-link-freetext" href="http://www.scyld.com/expert/wake-on-lan.html">http://www.scyld.com/expert/wake-on-lan.html</a>.<br><br>Unfortunately, I have been unable to wake up a node. Here is what I do:<br>"halt" a node. Detach the power cable. Reattach the power cable.<br>At this point the lights on the two onboard NICs (the Tyan web site<br>and the printing on the chips say that those are 3c920, the 3c59x driver<br>identifies them as 3c980; I don't know whether that is relevant; the NICs<br>work fine) come on. A Tyan technician told me that WOL on the Thunder K7 is<br>always on, no special BIOS setup would be needed. They also told me that I<br>have to use a 2.4.x kernel because only those would support APCI. I don't<br>understand why the kernel is important here: when the node is halted<br>what difference does the kernel make for the receiving of the magic WOL<br>packet that is supposed to wake up th
e box? Anyway, I compiled a<br>2.4.9-ac8 kernel with APIC enabled, which I use with the "noapic"<br>kernel option in /etc/lilo.conf. I have also tried the stock RH 7.1<br>2.4.3-12smp kernel without any difference with respect to WOL (i.e.,<br>no success).<br><br>After reattaching the power cable I then send the magic packet from<br>the master node:<br><br>./ether-wake -i eth4 00:E0:81:03:21:DD<br><br>where 00:E0:81:03:21:DD is the MAC address of one of the onboard NICs<br>on the node. tcpdump shows that the packet actually is sent. Also the<br>lights in the NICs on the sending and receiving end flash, but otherwise<br>nothing happens.<br><br>What's wrong? Any suggestions are most appreciated.<br><br>Thanks!<br><br>Martin<br><br>========================================================================<br>Martin Siegert<br>Academic Computing Services                        phone: (604) 291-4691<br>Simon Fraser University                            fax:   (604) 291-4242<br>Burnab
y, British Columbia                          email: <a class="moz-txt-link-abbreviated" href="mailto:siegert at sfu.ca">siegert at sfu.ca</a><br>Canada  V5A 1S6<br>========================================================================<br><br>_______________________________________________<br>Beowulf mailing list, <a class="moz-txt-link-abbreviated" href="mailto:Beowulf at beowulf.org">Beowulf at beowulf.org</a><br>To change your subscription (digest mode or unsubscribe) visit <a class="moz-txt-link-freetext" href="http://www.beowulf.org/mailman/listinfo/beowulf">http://www.beowulf.org/mailman/listinfo/beowulf</a><br><br></pre>
  </blockquote>
  <br>
  <div class="moz-signature">-- <br>
  <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
  <meta name="GENERATOR" content="Mozilla/4.72 [en] (X11; U; Linux 2.2.14-12smp i686) [Netscape]">
  <hr width="100%">
  <table cols="2" width="100%" nosave="">
    <tbody>
      <tr nosave="">
        <td nosave="">Claude Pignol</td>
        <td align="Center" nosave=""><a href="http://www.seismiccity.com">
SeismicCity, Inc.</a>
        </td>
      </tr>
      <tr nosave="">
        <td>2900 Wilcrest Dr.    Suite 470</td>
        <td align="Center" nosave=""> Houston TX 77042</td>
      </tr>
      <tr nosave="">
        <td>Phone:832 251 1471 Mob:281 703 2933</td>
        <td align="Center" nosave=""> Fax:832 251 0586</td>
      </tr>
    </tbody>
  </table>
  </div>
  <br>
  </body>
  </html>

--------------060806040101010806080706--

--__--__--

Message: 3
Date: Fri, 7 Sep 2001 21:39:08 -0400 (EDT)
From: "Robert G. Brown" <rgb at phy.duke.edu>
To: Greg Lindahl <lindahl at conservativecomputer.com>
Cc: <beowulf at beowulf.org>
Subject: Re: WOL: how does it work?

On Fri, 7 Sep 2001, Greg Lindahl wrote:

> On Fri, Sep 07, 2001 at 12:20:48PM -0700, Martin Siegert wrote:
>
> > Here is the problem: Each node draws a current of about 1.5A (I measured
> > that a few days ago). Since I have about 70 of those, booting all nodes
> > at once will draw all of the sudden a current of more than 100A. The
> > people who run our machine room don't allow me to do that (probably for
> > good reason).
>
> I can't really answer your question, but there's an alternate
> solution. You can use a device which delays the booting of some
> nodes. For example, the APC MasterSwitch has the ability to let you
> power cycle nodes by attaching to a web browser, but another feature
> is that it can power up the nodes with a delay after a power failure.
> It's a bit expensive for this purpose ($354 for 8 plugs @ 120V, 12A
> total), but maybe you can find something cheaper, such as an X10 based
> controller.

I haven't yet tried it, but a lot of ethernet cards now support Wake On
Lan, and ATX power supplies can boot in software once power is delivered
to the switching supply.  They are usually the better ethernet cards
anyway, the sort one would probably prefer to use in a cluster.

We were hoping/planning to arrange it so that a relative few master
nodes controlled when the slave nodes start up (and shut down in the
event of a loss of AC).  Is this not possible?

   rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu

--__--__--

Message: 4
Date: Sat, 8 Sep 2001 01:31:00 -0400 (EDT)
From: Donald Becker <becker at scyld.com>
To: Martin Siegert <siegert at sfu.ca>
cc: beowulf at beowulf.org
Subject: Re: WOL: how does it work?

On Fri, 7 Sep 2001, Martin Siegert wrote:

> I am trying to get wake-on-lan (WOL) to wirk on my Beowulf cluster
...
> Unfortunately, I have been unable to wake up a node. Here is what I do:
> "halt" a node. Detach the power cable. Reattach the power cable.
> At this point the lights on the two onboard NICs (the Tyan web site
> and the printing on the chips say that those are 3c920, the 3c59x driver
> identifies them as 3c980; I don't know whether that is relevant; the NICs
> work fine) come on.

This is likely a ethercard-specific problem.  Pre-CX 3Com cards don't
automatically go into wake-on-LAN mode.  The driver must first be
loaded, and the card left in the correct state (TotalReset + ACPI-D3).
My 3c59x.c driver takes care to do this properly.

I believe that the new 3c905CX cards do have a setting for automatically
configuring the card for WOL with just stand-by power.

Most other Ethernet adapters enable wake-on-magic-packet when stand-by
power is first applied.

> A Tyan technician told me that WOL on the Thunder K7 is
> always on, no special BIOS setup would be needed.

Likely true.  If you use a WOL cable, the ethernet adapter almost
literally pushes the power butter.  If you rely on standby power from
the PCI slot, the chipset must default to treating the PME signal as a
power-on signal.

> They also told me that I have to use a 2.4.x kernel because only those
> would support APCI.

That's false, and mostly not relevant.
My pci-scan code adds PCI power management state control to the 2.2
kernel, which is part of the ACPI spec.

The only aspect which is relavent is the ability to soft power down the
system.  That might require an ACPI Control Language interpreter if your
motherboard does not have APM functions.

> I don't understand why the kernel is important here: when the node is
> halted what difference does the kernel make for the receiving of the
> magic WOL packet that is supposed to wake up the box?

Yup.  After restoring power, the OS has never had a chance to run.  Only
the power-down procedure depends on the kernel support.

Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993

--__--__--

Message: 5
Date: Sat, 8 Sep 2001 01:31:58 -0400 (EDT)
From: Donald Becker <becker at scyld.com>
To: "Robert G. Brown" <rgb at phy.duke.edu>
cc: Beowulf List <beowulf at beowulf.org>
Subject: Re: WOL: how does it work?

On Fri, 7 Sep 2001, Robert G. Brown wrote:

> We were hoping/planning to arrange it so that a relative few master
> nodes controlled when the slave nodes start up (and shut down in the
> event of a loss of AC).  Is this not possible?

The Scyld Beowulf system has wake-up and sleep support built in.  

Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993

--__--__--

Message: 6
Reply-To: <LuismarP at cheque.uq.edu.au>
From: "Luismar Marques Porto" <LuismarP at cheque.uq.edu.au>
To: <Beowulf at beowulf.org>
Subject: Price of a 24 cluster
Date: Sat, 8 Sep 2001 18:08:29 +1000
charset="iso-8859-1"

Dear Beowulf Angels,

Is there anyone out there willing to help me estimating the
price of setting up a 24 PC beowulf cluster with Intel P4, 2.0GHz, 
dual CPU, 1 GB RAM, 4x40 HD/node, including switches and a bridge 
for a ATM conection? 

I just need an estimate at this point, but any particular 
configuration with respective amount spent would be of a really 
big help, since I am in a hurry to submit a resarch project.

If you prefer, you could just tell me about your experience sending
a particular message to luismarp at cheque.uq.edu.au.

I really appreciate any comments and help.

Regards,

Luismar

Luismar Marques Porto
Laboratory for Biological Engineering
Department of Chemical Engineering
The University of Queensland           
AUSTRALIA                   

--__--__--

Message: 7
Date: Sat, 8 Sep 2001 17:54:10 +0000 (UTC)
From: Martin WHEELER <mwheeler at startext.co.uk>
To: <beowulf at beowulf.org>
Subject: RE: D-Link switch for b at h

Thanks to all who responded to my query, both off-list and on.

Surprisingly (to me), my best-value-for-money turned out to be a Netgear
*sixteen* port switch, rather than any of the 8-port switches I had been
looking at.

Once again, thanks for pointing me in the right direction.

msw
-- 
              *** Free Speech *** Free Dmitry Sklyarov ***
   Sell your shares in Adobe.  Boycott ALL American non-free software.
                                 see:
http://uk.eurorights.org/                    http://uk.freesklyarov.org/

--__--__--

_______________________________________________
Beowulf mailing list
Beowulf at beowulf.org
http://www.beowulf.org/mailman/listinfo/beowulf

End of Beowulf Digest