PoempelFox Blog

Tue, 10. Feb 2009

MSI 945GSE Board arrived Created: 10.02.2009 23:23
Last modified: 18.02.2009 20:00

I finally got my MSI 945GSE-A board today. I have been looking for a proper Atom based mainboard for almost a year now. "Proper" in this context means:

at least 2 network interfaces onboard, at least one of which must be gigabit and from a good manufacturer
good chipset. That means: 945 GSE or US15. Especially NOT 945 GC
passive cooling, low power consumption

Although that really is not too much to ask for, up to now that was not really available.
For starters, most boards used the 945 GC chipset. That chipset is just horrible for this application - the complete idiocy of taking a processor that uses only 2 - 4 watts and coupling that with a desktop chipset that uses 22 watts and usually requires active cooling should be pretty obvious. In contrast, the 945 GSE only uses 6 watts.
And if you ever found a board that had two network interfaces, you could bet that it was either 100 mbit, or with a can't-handle-100-mbit-stable-leave-alone-gigabit chipset, or BOTH.
The MSI now does fulfill all those: As the name suggests, it uses the 945 GSE chipset, and thus can be passively cooled without any problems. It also has two nice Gigabit intel NICs onboard. Plus DVI, VGA, the usual audio outputs, and even 2 serial ports on the back and 4 additional port headers on the board. The only downside I have found so far is that the 1.6 GHz hyperthreading Atom N270 used on the board is NOT a 64 bit processor, i.e. it cannot execute AMD64/Intel64 operating systems.
Here are some pictures to make you jealous:
MSI 945GSE-A Mainboard top

The case is a Morex Cubid 2677 V3.0. Note that the two case fans visible on the third picture are not connected. The case fits the board pretty well, there is only one slight problem: The power LED has a 3 pin connector, while the board has only a 2 pin header for connecting it. You will have to work around that if you want the LED to light - or you can chose to just not care and do without the green LED.
The whole system has no moving parts except in the DVD ROM. Hard Disk is a 32 GB SSD. Power supply is in the front (below the DVD) and also fanless. It needs 12 volts / 5 amperes DC as input (supplied by an external fanless power supply) and generates the usual ATX voltages from that.
With the case closed, lmsensors shows a CPU temperature of 84 degrees peak when compiling the linux kernel. As fas as I know, the allowed limit is 99 degrees on-die - note however that the 84 degrees mentioned are not measured oncore, but through a sensor on the board. I have no idea how good that sensor is, but it seems as if the CPU is close to its limits - why they did not put a larger heatsink onto it when clearly they had the space beats me. Edit: My trusted source for processor electrical specifications tells me that the official limit is 90 degrees cover temp. That means temperatures are probably OK, as the board sensor usually is below the processor, where the heat dams up.

Installing Ubuntu Linux on the board proved a bit tricky. Although the hardware is all intel und thus very well supported by Linux, unfortunately the PCI IDs of the onboard NICs were not yet known by the e1000e driver in the installer of Ubuntu 8.04 (hardy, kernel 2.6.24) or 8.10 (intrepid, kernel 2.6.26). That makes a network install (as I usually do) a bit hard... I had to use the "jaunty" Beta for installation. It uses a 2.6.28 kernel and recognized the network interfaces without any problems.
A collegue tried Opensolaris version 2008.11 on the board: The LiveCD saw the network just fine, but did not manage to bring up X - that however might be a result of the really old 14 inch monitor connected at that time.

Here are some things I tried on the board, and whether it is fast enough or not. The CPU usage is taken from top and the sum of all relevant processes (e.g. mplayer and Xorg). Hyperthreading was enabled.

Play DVDs	Works fine. ~70 percent CPU.
Play highres videos (H.264 720p - try Big Buck Bunny)	Too slow (though not by much)
Play not-so-highres videos (H.264 480p - try Big Buck Bunny)	Works fine. ~80 percent CPU.
Play a DVB-S channel streamed over network (pal+, ~6500 KBit)	Works. ~100 percent CPU.
Pull 1 Gigabit from the network with `netcat $host $port >/dev/null`	Works fine (112 MB/s). ~50 percent CPU.
Push 1 Gigabit to the network with `cat /dev/zero \| netcat -l -p $port`	Works fine (100 MB/s). ~70 percent CPU for netcat and 30 percent for cat.

For measuring temperatures/voltages/fan speeds (should you really connect one), lm-sensors can be used: The board has sensors detected as Fintek F71882FG/F71883FG Super IO Sensors (linux kernel module f71882fg). In the same SuperIO chipset there also is a watchdog timer which can be enabled in the BIOS, but I have not found Linux-Support for it yet. Be careful when enabling it, it seems disabling it again is not that easy - only after setting timeout to 0, watchdog to disabled, saving, and then powercycling, it stopped resetting.

I measured (or rather: estimated with the help of electrical equipment) the power consumption for the whole system (dvdrom, ssd, board, 2 gb memory in the case with the external and internal power supply) with my cheap ammeter: It shows 15.7 watts when idle, and a peak load of 20.5 watts (with 2 burnMMX and glxgears).

For reference, here is the lspci -nn output from the machine:

00:00.0 Host bridge [0600]: Intel Corporation Mobile 945GME Express Memory Controller Hub [8086:27ac] (rev 03) 00:02.0 VGA compatible controller [0300]: Intel Corporation Mobile 945GME Express Integrated Graphics Controller [8086:27ae] (rev 03) 00:02.1 Display controller [0380]: Intel Corporation Mobile 945GM/GMS/GME, 943/940GML Express Integrated Graphics Controller [8086:27a6] (rev 03) 00:1b.0 Audio device [0403]: Intel Corporation 82801G (ICH7 Family) High Definition Audio Controller [8086:27d8] (rev 02) 00:1c.0 PCI bridge [0604]: Intel Corporation 82801G (ICH7 Family) PCI Express Port 1 [8086:27d0] (rev 02) 00:1c.1 PCI bridge [0604]: Intel Corporation 82801G (ICH7 Family) PCI Express Port 2 [8086:27d2] (rev 02) 00:1d.0 USB Controller [0c03]: Intel Corporation 82801G (ICH7 Family) USB UHCI Controller #1 [8086:27c8] (rev 02) 00:1d.1 USB Controller [0c03]: Intel Corporation 82801G (ICH7 Family) USB UHCI Controller #2 [8086:27c9] (rev 02) 00:1d.2 USB Controller [0c03]: Intel Corporation 82801G (ICH7 Family) USB UHCI Controller #3 [8086:27ca] (rev 02) 00:1d.3 USB Controller [0c03]: Intel Corporation 82801G (ICH7 Family) USB UHCI Controller #4 [8086:27cb] (rev 02) 00:1d.7 USB Controller [0c03]: Intel Corporation 82801G (ICH7 Family) USB2 EHCI Controller [8086:27cc] (rev 02) 00:1e.0 PCI bridge [0604]: Intel Corporation 82801 Mobile PCI Bridge [8086:2448] (rev e2) 00:1f.0 ISA bridge [0601]: Intel Corporation 82801GBM (ICH7-M) LPC Interface Bridge [8086:27b9] (rev 02) 00:1f.1 IDE interface [0101]: Intel Corporation 82801G (ICH7 Family) IDE Controller [8086:27df] (rev 02) 00:1f.2 IDE interface [0101]: Intel Corporation 82801GBM/GHM (ICH7 Family) SATA IDE Controller [8086:27c4] (rev 02) 00:1f.3 SMBus [0c05]: Intel Corporation 82801G (ICH7 Family) SMBus Controller [8086:27da] (rev 02) 01:00.0 Ethernet controller [0200]: Intel Corporation 82574L Gigabit Network Connection [8086:10d3] 02:00.0 Ethernet controller [0200]: Intel Corporation 82574L Gigabit Network Connection [8086:10d3]

3 comments

quote: Push 1 Gigabit to the network with cat /dev/zero \| netcat -l -p $port Works fine (100 MB/s). ~70 percent CPU for netcat and 30 percent for cat. Well, if cat takes 30%, why not leave it out and just use "netcat -l -p $port < /dev/zero"? Stdin redirection for the win...
htlaschi	23.02.2009 14:14

Yes, of course that would have saved CPU and given the same 112 MB/s that were achieved in the other direction. But then again, with that I could not have earned a "useless use of cat" award.
Fox	23.02.2009 18:34

Geht das auch auf deutsch? Ich verstehe kaum ein Wort.
bsdice	13.12.2009 22:48

write a new comment:

Sun, 08. Feb 2009

Fun with management cards

Created: 08.02.2009 01:33

Practically every server today comes with some sort of onboard management interface. Through that interface, the server can - via network of course - be remotely powered up, down or reset if it has crashed. A serial console is practically always available too, which is everything you need for linux or other proper (unixoid) operating systems. If you think you need to run servers on windows, you'll have to get one of the really good ones (and sometimes pay a bit extra for additional licenses), that allow you to redirect graphical output, mouse and keyboard to your pc (usually through a java applet). As an added bonus, they usually allow you to insert an .iso file on your local hard drive into a virtual cdrom drive on the server, so that you e.g. can boot from it.
However, all major manufacturers have in the past managed or still manage to mess some things up in these management interfaces, so here is my personal list of favourite screwups:

One manufacturer, lets just say the name starts with an S., manages to build management cards onboard that take 5 minutes to boot. Now that alone would not be horrible - but the thing is, you cannot even power on the server before the management has fully booted! That's right, you plug in the power, and maybe 5 minutes later you're allowed to actually turn the machine on.
There is another manufacturer, lets just call him Howling Packrat. Of course, any similarities in name are pure coincidence. The remote control java applet of this one used to be a bit delicate, it would only run with one exact version of java. Not 0.0.1 higher or 0.0.1 lower. Just that one version. Of course, it will happen that you try to run the applet on the wrong version of java, usually resulting in the applet locking up. So you kill your browser and try again. However, you don't really have many attempts there, because usually the management card will lock up after the applet crashed 1 or 2 times, and you have to power cycle the server physically to bring it back to life.
A different manufacturer of business machinery, active internationally, does not seem to have understood the concept of TCP on their management interfaces yet. I noticed this last week, when I couldn't get the Java Applet for graphical redirection to run. It just kept spewing nonsense errors (permission denied, class not found, ...) at me. It took me a while to realize that sometimes the simple webpages from the webinterface just stopped loading in the middle of the text. From there on, it was pretty clear what happened: In the management network there was still one 10 MBit switch, because there is close to zero traffic on this network and thus nobody ever cared to exchange it. All remote managements we had ever used before had no problem with that, but not this one: Instead of slowing down when the acknowledgements from the receiving side don't come, and resubmitting packets that got lost due to overflows on the slow link, as a proper TCP implementation would do, this one just closes the connection. That caused the webinterface-pages to be incomplete sometimes, and it prevented the Java applet from ever loading successfully, because the transmission of the rather big file would always die after a few kilobytes. As soon as the switch was exchanged with a 100 MBit one, all of these problems dissapeared. I wonder what will happen if that manufacturer ever builds management cards with a gigabit interface... Hopefully they will have learnt how TCP works by then.

no comments yet

write a new comment:

Sun, 01. Feb 2009

Musikdownloadversuche beim Bloed-Markt

Created: 01.02.2009 11:12

Letzte Woche sah ich bei heise.de dass der (ich-bin-doch-nicht-)bloed-markt sein Angebot von dem in Deutschland leider immer noch ueblichen DRM-Geficke auf reines MP3 umstellt, und bis 9. Februar guenstige Einfuehrungsangebote hat: 300 Musiktitel fuer 50 cent pro Stueck, und 300 Alben fuer 5 Euro das Stueck.
Es freut mich, dass es sich anscheinend endlich auch bis zu den hirntoten Managern der Musikindustrie herumgesprochen hat, dass es keine gute Idee ist, die zahlende Kundschaft mit DRM zu quaelen bis auch der letzte ehrliche Kaeufer vergrault ist.
Ausserdem finde ich den Preis von 50 cent fuer ein gutes Lied durchaus fair - anders als die normalen Preise, die ich als ueberzogen bezeichnen wuerde.
Das Angebot war also so interessant, dass ich es ausprobieren wollte. Zuerst positive Ueberraschung: sogar das Probehoeren (30 sec. Ausschnitt aus jedem Lied) klappt mit meinem Ubuntu problemlos.
Also mal testweise zwei Stuecke geklickt und fuer 1 Euro gekauft. Oder auch nicht: Zur Bezahlung standen zur Auswahl:

ein ominoeser Anbieter bei dem man sich registrieren muss, und ueber den man dann bei ganz vielen Shops ganz toll zahlen kann. Das ganze klang zu sehr nach PayPal um mir sympathisch zu sein.
mit Musikdownload-Coupons, die man im bloed-markt kaufen kann. Hatte ich natuerlich nicht.
per Kreditkarte

Na aber hallo, Kreditkarte klingt doch gut, auch wenn ich mich frage wie sie da mit Gewinn verkaufen wollen - die Kreditkartengebuehr muss doch bei so Pfennigbetraegen hoeher sein als der Rechnungsbetrag...
Bei dieser Option wird die Abrechnung ueber worldpay durchgefuehrt, man landet also auf einer worldpay-Seite - und nach dem ersten "Weiter" Klick wurde ich von einer Fehlermeldung begruesst. Wirklich klasse. Wenigstens gibt die Fehlermeldung einen Anhaltspunkt wo es hakt: Man muss im Firefox third-party-cookies akzeptieren lassen. Danach liess sich die Bezahlung durchklicken - um auf der letzten Seite ein Popup (natuerlich von Firefox zuerst mal geblockt) mit einer absolut nichtssagenden Fehlermeldung zu liefern.
Geklappt hat die Bezahlung aber wohl trotzdem, denn es kam eine Bestaetigung per Mail, mein Warenkorb war ploetzlich leer, und unter Downloads fand ich die 2 erworbenen Lieder - eines davon allerdings doppelt gelistet. Die MP3-Dateien inclusive cover-jpeg (in Briefmarkengroesse) liessen sich problemlos als .zip runterladen, die doppelt gelistete Datei war allerdings auch doppelt enthalten, einmal mit angehaengter (4) im Dateinamen. Egal, es hatte geklappt, ich war zufrieden.
Die Zufriedenheit verschwand ganz schnell als ich es am naechsten Tag wieder probierte, diesmal mit mehr Liedern: Offenbar hatte der Server Alzheimer, denn bei praktisch jedem Klick wechselte mein Zustand zwischen "eingelogt" und "nicht eingelogt", "Warenkorb voll" und "Warenkorb leer", hin und her. Ausserdem hatte er meinen "Merkzettel" vergessen.
Sonntag frueh klappte es dann wieder problemlos. Da wurde mir dann auch klar wieso das eine Lied doppelt war: Um auf die Seite zu kommen, auf der man alles zusammen als .zip runterladen kann, muss man bei irgendeinem Lied auf "Redownload anfordern" klicken. Dadurch wird genau dieses Lied aber doppelt in das .zip gepackt, und als Bonus: Man verbraucht damit auch 2 der maximal 3 Downloadversuche die man pro Lied hat, obwohl man ja ueberhaupt nur eine einzige Datei, naemlich das .zip, herunterlaedt.
Fazit: Preise ok, aber leider nicht dauerhaft und fuer alles so. Shop vor dem Kauf benutzbar. Bezahlverfahren ein einziger Graus, Gefrickel hoch zehn. Und wenn man erstmal bezahlt hat, ist man ganz offensichtlich nicht mehr wichtig, denn die Bedienbarkeit des Downloadportals laesst stark zu wuenschen uebrig und hat offensichtlich Bugs. Sicher nichts was ich regelmaessig benutzen werde, dazu muessten sie erstmal ihren Webauftritt von jemandem ueberarbeiten lassen der sich mit sowas auskennt, und dauerhaft vernuenftige Preise bieten.

no comments yet

write a new comment:

EOPage - generated with blosxom