X64, buffered memory, and power supply

Hello community,

I've had an interesting thing happen today. If anyone can explain it, I'll be very grateful.

So there I was, installing OpenWrt on an x64 device. Specifically, it was a Sophos SG 105 rev 2 (Intel Atom E3826, 2 GB RAM, 64 GB SSD). Doing post-installation tasks and noticing that buffered memory count is creeping up on a nearly-idle router. Did a reboot and watched buffered memory count rise to 1.6 GB within minutes.

This is the first time I've noticed it on any x86 device, and I've dealt with quite a few, including nearly identical Sophos units. So I asked myself: what, if anything, is different about this install? The only difference I could think of was, I got lazy and instead of grabbing a 12V / 3A power supply from the shelf (the router is rated for 12V / 3A), I hooked the router up to a 12V / 5A power supply that was already on the workbench. So power down, swap the power supplies, power up, and... buffered memory is at 12 MB after 30 minutes of uptime.

So... what the heck did I just observe??? The inquiring minds want to know... :slight_smile:

Keep in mind the 5A power supply doesn't push more current. It can supply up to 5A. You usually don't want a power supply at 100% anyway with 40-80% being the sweet spot.

Maybe check the voltage to make sure it's within spec?

1 Like

Unfortunately, I am not equipped for electrical measurements. But, assuming the PSU is under or over nominal voltage, how would that lead to the ballooning of buffered memory? What's the mechanism?

That's how it's supposed to be, buffer/cached mem high, free mem low.

1 Like

I understand that, but what I don't understand is why the pattern of bufferization changed when power supply was replaced... What is it about power that drives bufferization (or lack thereof)?

More than likely, the memory situation was coincidental. Remember, correlation does not equal causation.

If you really want to prove that there is a causal link, try putting the old power supply back into service, but make sure you have full control of the network traffic during the test (maybe take all clients offline except for one, run some speed tests, web browsing, video streaming, etc.). Then swap to the larger 5A supply and run the same tests... watch the memory and see what happens.