And some other interesting things transpire to interfere with testing this. Kingston appears to have suffered some trouble with their KVR16E11K4/32. As far as we've been able to determine, these kits are available in three variants: HYN (Hynix), ELP (Elpida), and KTC (Kingston Tech Corp). Alas the set that we had in inventory at first was KTC, which showed some disturbing and interesting issues, and prompted ordering a second kit, which was an Elpida kit, which turned out to be a total fail. At that point I ended up swapping out an older Sandy Bridge era set of 1333 modules to validate that it wasn't the ASRock board; that worked like a champ. There's been a lot of problems with Kingston and Supermicro Haswell boards, and this seems like a similar issue. Regardless, I plowed on, did a little research, identified that the Hynix version was supposed to work for Haswell, later confirmed by Kingston (see review response), put in yet another requisition, and managed to obtain a set of Hynix.

I apologize to anyone who has been waiting for updates on this evaluation; requisitioning hardware typically takes at least two weeks to complete the paperwork, ordering, waiting for ground shipping, and then to find another timeslot to spend some time on this... and that had to happen twice! But since the Kingston parts are having trouble on Supermicro boards too, I do not feel that this reflects poorly on the ASRock board, but rather it reflects poorly on Kingston.

So, onward. Login!

High end servers have historically come with out-of band (also known as "lights out remote management") systems, which are basically a microcontroller residing alongside the server which provides management services. In the old days, we had serial consoles and power distribution units that could cycle each outlet on and off, then slowly things like the and Dell DRAC appeared, then HP's rapidly evolving iLO line. Along the way, we gained features big and small, such the ability to flash a Unit ID LED (UID), to help identify a unit in the rack, and virtual media, to allow the remote "insertion" of virtual USB devices to allow the server to be loaded with an OS or tools, and remote reviewing of the system event log.

For whatever reason, I hadn't actually seen American Megatrends' Megarac SP in detail, possibly because these last few years had been heavily HP, IBM, and Supermicro-oriented (Dell is still on the vendor badlist). But I had reasonable expectations based on experience with other kit. I pulled up the login page. And then I knew this was going to be fun.

"Required Browser Settings". What? It TESTS? And American Megatrends hits a home run for best first impression of an IPMI BMC implementation ... because every sysadmin I know is paranoid, and runs into problems with browser settings like this all the time, and it is always annoying to play the game of "what'd I forget now."

Someone actually had some professional sysadmins try this during the development phase, I'd say.

OK, fixed. Login!

And this is gorgeous. It is simple, it is organized. It gives some impression that someone took some time and care to do little things, like knowing that someone doing maintenance on a server would probably like the UID status right there on the home page, though interestingly it seems that the UID light on the chassis is ON when it is OFF in the IPMI... which was fixed after a reboot.

Turns out that if you power on the server and hit the UID button on the unit quickly thereafter, while Megarac SP is booting, you can get the Web GUI and the hardware 180 degrees out of phase. This might come as a bit of an unpleasant shock to the rack and stack guys who in some cases are installing servers, powering them on, and letting some remote tech know the MAC address.

I am guessing that this is not an ASRock problem per se, but if they talk to American Megatrends about it, it would be nice to have a server power indicator added to the home page, and maybe a UID and power indicator added to the login page too.

Server Health, Sensor Readings

IPMI is sufficiently well developed at this point that sensor readings are nonoptional. Regardless, the treatment is nice, they're logically sorted, and there's a graph of event log entries.

I'm not sure how actually USEFUL that is, but while clicking around I noticed

Which leads on over to the event log section if you click through.

That strikes me as a bit odd as well, since the shop is about sixty five degrees F, and it was here on 02/14/2014. It does seem to have chalked up a large number of UNC events prior to arriving here as well.

So there are all the typical configuration objects you would expect, but a few interesting ones too.

This is a Work In Progress. Please check back.