Tagamet, we got an answer back from engineering. The system has to "think" that it only had one power supply to begin with. So are you ready for a little downtime ?

You'll have to completely remove power from the system and power it up with only a single power supply inserted. Then there is no redundancy to "lose". I just tested this on one of our 1850's in the lab. I had removed the 2nd supply, got the red-x and amber light on front of the server. I completely shut it down and removed the power cords, pulled out the 2nd power supply and turned it back on. Now it only has blue light and no errors.

If you re-insert the 2nd power supply (without power cord) - it will go into error state and you would have to repeat process all over to return to original state.

I think better solution is to do the Y-cables, now that you know it is not a doubling in power, only ~17%. Even though this won't protect you from a power leg failure, you can at least sustain a power supply failure. And you won't have to deal with this workaround, and you could perform the Y-cable upgrade without taking the machines down.

What's your userid on the wiki ? Couldn't find you, and this is the only logical place I could think to contact you.

Hi Scott,
Excellent! Thanks a million for your research and followup with engineering
I had not signed up on this wiki until just now, so thanks for posting this here


Ok, since our mgmt felt that the additional 17% power hit on the y-cable solution was too much, I'll proceed with the 2nd solution (power off, unplug 2nd psu, power on).

Btw, this explains why our newer 1950s don't exhibit the same red x errors in ITA. When they were racked and powered on, they already had their 2nd psu unseated from the chassis. (whereas our 1850s had both PSUs seated when powered on)
We just happen to have a downtime coming up very shortly so I'll perform this powerdown/unplug-psu method


Thanks again for your hard work. You just saved me a lot of headaches
-Tagamet

Glad to be of assistance. Hopefully see you on the chat tomorrow. Todd's going to talk about virtualizing enterprise apps and probably have some additional Dell Engineers on as well.

Scott,

I tried out your procedure on a single PE 1850 and it worked like a charm
I'll be doing this on all our poweredge servers during our next downtime. Thanks so much for your solution

-T

I guess I never really thought about this, but I am assuming by your remarks Scott, that it would be fine to have each power supply conencted to a seperate power source?

Best practice would be to have each power supply going to a separate leg of power. So in your datacenter if you have A and B power, you would want one power supply on A and the other on B. So if datacenter power leg A went down, server would stay up.

Typically the power legs are coming from different site wide UPSes, so you just make sure that each power supply is connected to different UPSes.

The "Y" cable solution was specific recommendation to tagamet's environment where they have some unique power retraints. That would not be the recommended solution in most customer's cases.

Make sense ? Thanks for joining in the converation.

Thanks Scott,

I do have enough backup power on seperate feeds to do this. I just never thought about it.

FYI, I did a more thorough analysis of the power differences across multiple servers, and multiple workloads.

http://www.delltechcenter.com/page/Cost+savings+through+removing+server+power+supplies

Excellent research Scott & team - thanks a lot
Very interesting to read your results. Your mix of servers are exactly what we have in our DC (except for the newer 1950s

I'll share your results w/ our DC OPs/Engineering/Mgmt

-T