Replication and power consumption?

[FrankWard]

My Scale Master and Core Slave servers together pull between 140W-230W as reported on my UPS. The idle load jumped to 230W soon after a replication task kicked off and persisted until many hours AFTER the replication event finished. Now it's back to 140W idle. The Scale server had no spikes or jumps in power or load, and from what I can tell, the Core server CPU increased from 36 to 56 degrees for the duration of the replication event and the drives were in use. This increase of 20% CPU and drive activity cost +90W power draw for the duration.

Am I on the right track? Is this normal? Anyone have thoughts or information to share on the topic?

 

[jenksdrummer]

Not to sound too sharp with this response, but if you're using the systems they do increase their power consumption. Depending on the replication, the data could be getting decompressed and recompressed; which will add a lot of overhead. Even if it's not, there's still the CPU deciding where to put the data; the NICs processing the packets, the disk arms placing the data; or burning it to SSD Cells...RAM processing / churning; CPU cores going from parked/idle to performing work...lots of it adds up.

 

[jenksdrummer]

FWIW, it's a good thing that you are seeing it drop load when idle - means it's doing what it can to save you money / power / thermals ;)

 

[FrankWard]

Not to sound too sharp with this response, but if you're using the systems they do increase their power consumption. Depending on the replication, the data could be getting decompressed and recompressed; which will add a lot of overhead. Even if it's not, there's still the CPU deciding where to put the data; the NICs processing the packets, the disk arms placing the data; or burning it to SSD Cells...RAM processing / churning; CPU cores going from parked/idle to performing work...lots of it adds up.

Of course. It appears that 20% more CPU = +90W draw on the UPS decreasing the runtime of the UPS by 41%. Couple this with the fact that you don't want to run your UPS battery lower than 30%, suggested 50%, and the runtime is lower or halved. This is based on ONE of the two servers actively working. If both are working it's even lower. So during a power loss event:

Runtime is:
~53 minutes @ 140W, both servers idle.
~28 minutes @ 230W if one server is actively working.
~14 minutes @ 430W if both servers are actively working.
~10 minutes to avoid discharging the batteries below 30%.
~7 minutes if shutting down to avoid discharging the batteries below 50%.