Powering Down DS25 Alpha Server

[Wim Van den Wyngaert]

Over here they do a power cut every 2 weeks and between 10 and 20 Alphastations go off and on again a second later. This for more than 10 years. Not had any problems yet except settings lost due to empty battery. Off course, these stations don't cool off as would your server (1s against a night). I would think that a device that never cools off is more reliable. But I'm not an electrician.

I have seen some disk failures after power off.

BTW : if you use shadowing it is very well possible that you will get a shadow copy if not properly shutdown. And on my systems it seems impossible to avoid completely (so test it).

Wim

[Wim Van den Wyngaert]

BTW2 : make sure that your monitoring tools detect things that fail during power on. Many tools use device errors or shadow set changes to detect them.

BTW3 : if a disk or whatever fails you will notice it in the morning at boot time. If running 24/24 you will notice it much sooner (if you have appropriate monitoring) and you can repair it before the business day starts.

BTW4 : what is the cost of doing the manual power off/on, the risk that the person is ill/late, the risk that boot goes wrong, ...

BTW5 : I like the green thought of it ...

Wim

[Thomas Ritter]

Michelle, did you find the advise useful ? Worthy of any points ?

[Robert Gezelter]

Michelle,

I will readily admit that I have not done this experiment on a full basis with the DS25, but once upon a time, I had an equivalent experiment occur naturally with a group of Professional 350 systems.

For many years, a Professional 350 served as my administrative system. As a matter of routine, I left it powered up and running 24x7x366. Failures were infrequent.

A client of mine (a law office, to be precise) had a pair of equivalent systems (all of these systems were originally purchased from the same source). As a matter of course, they powered down their systems at the end of each workday.

Neither of the two offices, which were within one mile of each other, had power conditioning or UPS systems, beyond perhaps an ISOBAR in each location.

The incidence of system hardwware malfunctions was significantly higher for the systems that were power cycled on an ongoing basis.

There is some basis for this. Drastic temperature changes does produce problems of expansion and contraction on components. Power cycling also often produces power surges every time the system is started. If the system is well designed, the impact is less, however cycling is certainly more stress on the system than a continuous state. Electromechanical components are also often responsible for more of the power consumption (and are also often less damaged by power cycling) than are VLSI components.

That said, idle shutoffs for spinning storage, good. Cycling server CPUs, less good. Check where you are actually burning power, it is often worth the effort.

- Bob Gezelter, http://www.rlgsc.com