The machine itself. I ordered a
bare-bones system with
no memory, no disc, and no operating system.
Physical dimensions: 187mm wide, 160mm front to back, 31mm high.
Mass: 1072g plus 159g for disc and memory, total 1231g
Power brick with USA power cord (interchangeable per locale). input 100-240 VAC 50-60Hz up to 1.5A. Output 12 VDC up to 5A. Power cord length is 60cm; the wire from the brick to the machine is 155cm, plus 11cm for the brick gives a total length of 226cm.
HDMI (male full size) to DVI-I female, can't tell if it's dual link but the holes are there. DVI-I has lines for analog output.
3.5mm male 4 wire plug to RCA (2 wire) connector, about 140cm long, for S/PDIF. I suspect that the 3rd wire on the 3.5mm plug is for S/PDIF.
Two Wi-fi antennae with male SMC connectors.
Packet of bottom feet and screws for mounting the disc.
Users Manual (English)
Steps in assembling the machine:
I put a piece of polypropylene foam packing material on the table and aluminum foil over it, and the machine upside down on top of that. I kept my wrist or arm always in contact with the foil, to minimize ESD (electrostatic discharge).
One Phillips screw holds closed the bottom access panel. Remove (and don't lose) the screw. Slide the panel to the right, towards the screw hole, about 1cm, then lift it out and turn it over.
On the right side the memory sockets are revealed. Be very careful of static discharge when handling memory. When removing the SODIMMs from their container, do not put force on the capacitors that form a border along the top edge of the circuit board, both sides. With the Kingston product there is a multilingual instruction sheet underneath the boards, with pictures. The boards go in at an angle and then you tilt them down flat so the grabbers can hold them.
The access door functions also to hold the disc. The socket end goes away from the screw hole and the disc's label should be up, i.e. away from the door panel. Attach the disc to the side rails with the four provided screws.
Replace the door and slide it to the left so the disc's power and SATA sockets connect, resulting in the closing screw holes lining up. Fasten the screw.
On the shell there are four round depressions where the feet go. The feet are on one sheet of plastic; use scissors to cut them apart through the flat area between them. For each foot, peel off the white paper backing. Position it so the raised part is centered over the depression, and press down. Hold down the foot and pull up a corner of the flat part, and it will tear in a circle around the foot.
To get into setup, turn off power, then hold down F2 on the keyboard and turn the machine on again.
I didn't change anything in BIOS setup, but here are some tidbits of information:
I experimented with Quick Boot, but all it seems to do is to not show the splash screen, which is on for at most 2 seconds, so messing with this is not worth it.
Partitions on the old Diamond: Total size 160Gb.
|3||10.8GB||primary||ext4||type=83, boot||Linux root|
On the new machine: Total size 500Gb.
I ran Memtest86+ from the installer disc overnight, total of
16 hours, 10 passes, no errors reported. The one anomaly was that it
RAM: 1032MHz (DDR3-2065).
With the USB fan on high (a little noisy), the case was only slightly warm to the touch.
I rebooted the install disc and started the rescue system, to get the MAC addresses of the two NICs. Without these configured, my internal firewall will refuse all packets from the interloping hacker. The two NICs are:
These are the numbers that the installer uses, but the production kernel may or may not reverse them. You can plug the network cable into either, and if on one of them it times out getting a DHCP address, it will try the other. It tries eth1 first.
I rebooted the installer and did a network installation. Setting it up took about 15 mins, mostly defining partitions. The installation itself took 4 minutes -- sneeze and you'll miss it.
After rebooting into the new system it did automatic configuration, then decided to shut down the network. Rebooting didn't help. The problem turned out to be /etc/udev/rules.d/70-persistent-net.rules ; the installer called the Intel NIC eth1, while the production kernel called it eth0. The wire was connected to the NIC that was down. I configured it by hand to match what the production kernel does. Then I ran yast2 network settings, and gave eth0 its static IPv4 address. It did RFC 2462 auto configuration for its IPv6 address, which is how I handle it on my net.
I need to remember to reconfigure the net to a bridge, for running the virtual machine.
Confirmed that I can use ssh to execute commands on Orion (password required).
I copied Diamond's /m1/custom/extra.sel onto Orion, to get special software that it needs.
Now I run post_jump. There were some wanted packages that did not get installed due to a ridiculous dependency problem.
Checkout script: all tests passed.
There is a worrisome symptom: you can't turn off the machine. If you use
normal means to shut it down, e.g. the
poweroff command, it will turn
off power, but after about 5 seconds it will boot again. Temporary
workaround: in the Grub menu press any key, like the down-arrow key, to stop
the auto-boot timer. Now unplug the power, hiss, boo.
The same thing happens if you suspend to RAM or disk, which is going to be a big problem for me since I want to put the machine to sleep when it isn't being used.
It turns out that the Intense PC uses a rebranded Intel DH77EB motherboard, and there is a bad interaction between Wake on LAN and suspend/shutdown. See this thread about reboot instead of shutdown, OP is gbarrios (2012-09-29), responses from Susan Sharp of Intel. gbarrios got relief by turning off WOL (not a good solution for me), and also by upgrading to the latest BIOS version, Intel's v00097.
See these instructions to update the BIOS. Basically you are going to prepare a bootable USB storage (flash) device with MS-DOS on it, download the new BIOS onto it, boot it, and execute the flasher.
I updated the BIOS successfully, and now I can suspend to RAM and it will stay suspended. However, it still boots 5 seconds after power off. There was an issue with the Fit-PC-2 in which there was a factory-soldered jumper; if absent the machine would stay off if power went off and then returned, or with the jumper it would boot automatically. Later variants all do the auto boot thing. I suspect that's what I'm seeing: power is on, the machine isn't, and so it boots. That's not a bug, that's a feature! I'll just have to live with it.
In fact, if you unplug the machine, then replug power, it will boot automatically according to its plan.
These issues remain to be set up, that are not covered in the checkout script:
Quite a number of wanted packages could not be installed. I have a feeling that many of them are back-version packages that should be updated from SBS, perhaps by an open dist-upgrade. This was done and the downloaded packages were saved on the distro server. Now the only wanted but missing packages are the ones that are actually unavailable anywhere, and no installed but unwanted packages remain either.
The printer presently is hosted on Diamond. Move it to Orion and deactivate the print server on Diamond. This requires the Canon printer driver to be installed. Leave the USB cable extender in place, since Orion will soon migrate to Diamond's present physical location.
Reconfigure Orion's eth0 as a member of a bridge, that KVM can use. [Done.]
Verify that KVM is working on Orion, by moving Petra (virtual) from Aurora onto Orion. [It works.]
Define the virtual machine for Baobei.
Move Baobei's disc (virtual) from Diamond to Orion. Try these alternative plans:
Make sure that Windows will boot in its new home.
Move the finance
partition to Orion and deliver it as a
disc. On the old Diamond it is a network share, causing no end of
complaints from Microsoft Office about possible virus infestation.
The webserver is needed for distro storage. Copy from Diamond. Set up HTTPS, needed for the future accounting server. [Done.]
For the backup role, these items need to be taken care of:
Copy the backup directory to Orion. [Done.]
How to do a bulk transfer from Jacinth: we can use port 1443 on Orion.
On Orion: netcat -l -n 1443 | (cd /home/backup && tar xpf -)
On Jacinth: (cd /home/backup ; tar cf - .) | netcat orion 1443
For checksum: find $dir/. -type f -print | sort | tr '\012' '\0' | xargs -0 -n 50 cat | md5sum
Change the CNAME of backup.cft.ca.us to point to Orion. [Done.]
Go through the whole backup process and make sure it works. [Done.]
Verify that the burner burns and checks the disc and that Xena can read it. [Done.]
The role of distro storage includes these items:
Create a new CNAME distro.cft.ca.us pointing to Orion. [Done.]
Copy the distro and post_jump directories to Orion. Distro goes in /s1/SuSE while post_jump is in /home. [Done.]
Set up symlinks to the distro directories in /home/httpd/htdocs (and remove from Jacinth). [Done.]
Move the distro-maint crontab off Jacinth onto Orion. [Done.]
Edit and install new repo definitions referring to distro.cft.ca.us
in the base URL. Put the definitions in the post_jump directory, for
both versions 12.3 and 11.4, in $PJ/$VERS/etc/zypp/repos.d/template.
For each host, executing on Orion, do
audit-repos -v -u -i
HOSTNAME. It will copy the repos from the above directory onto the
target, remove the old repo definitions and metadata, put the new ones
in place, and (with -u) re-download the metadata. [Done.]
When the machine is not running and power is on, the machine will boot. That's not a bug, that's a feature, intended to recover from a power failure. But if you really want the machine to be off, you need to unplug it. Shut it down and wait 5 seconds for it to boot. In the grub menu press any key to stop the auto boot timer. Then unplug power.
Suspend to RAM (S3): It sleeps within 1 second and wakes within 4 seconds. A LAN magic packet, a keyboard action, a mouse click, or the realtime clock will wake it. But on an optical mouse, motion is ineffective because the lamp is off. The best key to press for waking is Shift, since printing keys or mouse clicks may or may not be sent to the application.
It does not need
Suspend to Disk (S4): It hibernates within about 7 seconds (this probably depends on how much memory the running processes are using). It wakes in 28 seconds. The same events wake it as for S3.
The resume (swap) device is specified in /etc/default/grub as one
of the kernel command line parameters in GRUB_CMDLINE_LINUX_DEFAULT,
and after you change it you need to run
/boot/grub2/grub.cfg. You must not specify the resume device as a
symbolic link; it must be the block device, e.g. RESUME=/dev/sda2 not
/dev/disk/by-label/diamond-swap. (Hiss, boo.) To suspend you can use
the -r option of s2disk, but there is no opportunity to override the
resume device when the initrd tries to load the saved image. It will
just sit there with no video and no keyboard response.
Soft and Hard Power Off (S5 and S6): The auto reboot feature precludes testing what else will wake the machine from these states.
Diamond is a slave directory server. Those services need to migrate to Orion.
Stop and disable directory services on Diamond:
Copy the respective databases and files from Diamond to Orion.
Reboot both machines. They should adopt their new hostnames.
Enable and start servers one at a time.
Install Diamond's greeter background, not the Mathnet generic one. [Done.]
The default session is TWM which is not installed. Improve this.
The login box needs a title that includes the hostname.
After the user gives the password it should show something like "Authenticating", not redisplay the password prompt.
XFCE icons are mostly nonexistent. No calculator. [Done.]
Tab bar needs to be moved to the top. [Done.]
New script to start the virtual machine, and an icon for it. [Done.]
Virtual Machine issues:
Decide how we're going to present the /home/finance directory to the VM. Methods tried:
Samba share from Diamond: Windows thinks it's a security problem. I'm trying to get away from this method.
libvirt disc of type
dir: It comes out as a VFAT
filesystem and cannot be read-write. Useless.
libvirt filesystem: You specify a host directory and a dir seen by the guest used in the (Linux) mount command. The problem is, content is accessed by qemu and mapping the host's and guest's users is not exactly simple. It would be useful if readonly and everything is world readable.
Image file with NTFS on a partition. This would be mounted readonly on the host so we can back it up, and read-write on the guest. This is how I used to do it in VirtualBox.
The VM was operational yesterday, but now it always drops into Startup Repair.
Sound and multimedia checkout. Particularly Youtube videos. [Sound and Youtube are working. Still need complete checkout.]
ldaputil does a double starttls when connecting to ldaps:// URI. This blocks ldapsync from running. [Fixed.]
cronj needs to detect any established TCP connectios and not sleep.
The definition of Baobei for VirtualBox is /home/alice/.VirtualBox/Machines/baobei/baobei.xml . /home/finance is exported to Baobei as \\vboxsrv\finance.
The power used by the machine generally matches what's in the
specifications. These were measured with a
power meter and are believed accurate to ± 1 watt, although I do
not have any independent calibration of the device.
|Power off||0 W|
|Booting||21 W||Maximum; power varies|
|Doing nothing||14 W|
|Fan low||add 1 W|
|Fan high||add 2 W||Fan high for these tests:|
|SpeedSHA (I/O)||18 W||Saturating the disc|
|SpeedSHA (CPU maxed)||21 W||One core in use|
|Wastetime X1||24 W||1 core, took 18.0 sec|
|Wastetime X2||28 W||2 cores, 18.0 sec|
|Wastetime X4||30 W||2 cores hyperthread, 32.4 sec|
|Wastetime X8||30 W||took 76 sec|
Wastetime is a simple CPU-only integer benchmark, configured in
this case to fit in the L3 cache. The two cores ran without slowing each
other down, as expected with no use of memory. When hyperthread was used, i.e.
4 simultaneous processes, they squeezed out about 10% more CPU power.
8 simultaneous processes suffered about 5% slowdown compared to 2 processes,
probably because their data didn't actually fit in the cache.
I studied cooling of this machine, since 30 watts natural convection seems
improbable. In the following table temperatures are in degrees C. The CPU
has a thermometer for each core, but they were always identical. The criterion
reached steady state was that the temperature did not change
for 5 minutes.
|Light web surfing||None||52||46|
|Wastetime 1X||None||57||46||10 second test|
|Wastetime 2X||None||58||46||Jumped up within 5 secs|
|Wastetime 2X||None||66||51||Steady state in 20 mins|
|Wastetime 2X||Low||62||50||Steady state in 14 mins|
|Wastetime 2X||High||53||44||Steady state in 17 mins|
For comparison: Dell Studio 15 laptop, 30 watts at idle: CPU 75C, disc 43 C.
The rotating disc is quiet but not silent. For truly silent operation you need a flash memory (solid state) disc. The fan on low is quieter than the disc. But on high it is significantly noisy and moves a lot of air.