SMT / hyperthreading

[ThijssjihT]

I want minimal energy usage. I choose my hardware appropriately. Relatively recent Intel celeron with required (by me) low powerd C-states.

Should I enable hyperthreading?

Hyperthreading will increase power usage, but might also lead to the CPU finish the task faster, and return to low power C-state earlier.

What will be more energy efficient? Quick high power HT/SMT for a short amount of time, or less power for a longer time?

Does anyone have experience or knowledge about this?

[WilliamsKnights]

OK.. so here is the thing. 

Yes. HT can add to power consumption. but as you state it can reduce the time taken to perform a task. That is if that task can take advantage of the hyperthreading. Not all tasks are suited to HT.

So it's 100% dependant on the type of load you are placing on the CPU, what other tasks are ongoing and how the code is written.

There is no single rule that fits them all.

 

There are other things going on that can have a significantly bigger fluctuation on power draw. 

Differing hard disks can have differing initial spin up power requirements. even same brand, same model, different firmware.

 

Same for power supplies. Same manufacturer, same model but different production run can impact power efficiency.

 

I work in IT for a company with over 40,000 servers. Power consumption is something we monitor actively on our non cloud servers. (Who cares about the cloud power costs?).

One of my specialities is Database systems (infra DBA not APP DBA). Systems under my management have over 140,000 CPU cores. We have HT enabled on around 75% of the estate.

We tried disabling HT to see if it had any power impact. It did. Overall things took longer and the overriding view was MORE power was consumed with HT disabled.

 

I also manage middleware teams. HT had a lesser impact as the apps were less parallelized. here Disabling HT has a VERY SMALL impact to power draw. Not enough for us to spend the time implementing changes to disable further.

 

The areas where we disabled HT and left disabled were relating to security over performance and/or power.

 

Hope that helps.

 

[ThijssjihT]

Thank you for your answer. It helps more than I hoped for in an answer. Your personal experience with testing HT makes it trustworthy.

I will be enabling HT. As I understand it, it will likely help reduce power in most situations, and when it doesn’t, it will be peanuts compared to other hardware.

I have management over 2 CPU cores. I guess having 2 doubled to 4 threads will likely be more impactful than 8 to 16 for example, because of the higher chance those threads will be utilized.

I did look at all aspects of my build. Single channel memory (1 stick uses less energy than 2, and it is not a bandwidth heavy load I think), initially emmc memory (but changed that to normal sata flash because of compatibility issues), older CPU, but recent enough for advanced C-states, efficient PSU in the sub 80 watt range. And I know the most power hungry are the HDD drives. I got some lower range NAS drives. They are efficient. Not the most efficient, but I guessed their firmware would make them better fit, in addition to longevity features of such disks. I’ve never looked at firmware versions though, and if they are upgradable. I’ve yet to look into configuring spin-up and spin-down as well, if they are user configurable.

[WilliamsKnights]

My experience is from datacenter usage. I have over 30 years experience in IT. Mainly supporting IT in the financial sector where IT expenditure is high.

My first hard disk was a 20MB model. In my day job I currently consume over 35PB. I have seen it all. Hard cards, Hard Disks, Direct attached storage, 30 rack storage vaults.

Take below with a pinch of salt. It's my thoughts only as very little is proven in this field.

 

So much has moved to flash these days. Tiered storage layers are used with most used data being on flash and less used moving to slower layers. I checked with 2 companies we use. (cannot use names here but they are leaders in there fieldsa), generally all spinning hard drives are left with manufacturer defaults and not configured to perform additional or custom spin down operations. There are some outlying use cases where they do this. this such as long term storage vaults where read and write operations are minimal and delays can be planned around in software and OS levels.

 

When we had storage in servers. Generally it was not a good idea to power down or spin down drives.  Drives normally store 2 values. Power on cycles and running hours. BOTH are used to gauge wear and tear. It was generally accepted practice to leave it up to the drive manufacturer to set this up in firmware. You would buy the best drive for your situation.

Manufacturers such as WD and Seagate have drives for differing uses. Often it's just a firmware change. Other times it can be componentry. All depending on planned usage.

A drive setup to spin up and down generally have lower warranties than those set to spin uninterrupted. Perhaps this could also be used as a gauge to what's likely to be better.

So for a hard disk. It's a mechanical item. It takes high power to spin up but low power to maintain it spinning. Even a warm 'at temperature' bearing takes less energy to spin than a cold one. But the energy levels are tiny. 

Some people use a calculation such as 1 hrs spinning = 1 spin up cycle but this has never been proven as far as I'm aware. There are to many other factors to consider. Temperature, even elevation all have an impact. Also. If you have SMART features enabled. this can force a spin up cycle when SMART stats are updated. Even TrueNAS will force a spin up every now and then to check on drives.

 

There is no hard and fast rule. The most important thing is to look at your usage and make a decision.

If you are setting up a plex server that will only be used on weekends. then there is an argument that spinning down for 5 days is fine.

If your usage is you will be using 10-50 times a day then leave it spinning etc.

 

 

My Personal Bias

I used to focus on running costs at home. All my servers had power meters. I even got into the rabbit hole of how much energy a philips hue light used when turned off. (this is shockingly higher than expected, especially when you have all lights converted to smart lights).

I used to have a HP N54L packed with hard disks. It was just a plex server and I was aggressive with spin down cycles.

I purchased Seagate 3TB Barracuda drives as at the time they were best £ per TB. They had a 4 Year warranty.

I lost 2 of the 4x 3TB Seagate drives within 5 years. Seagate warranty had expired.  Was this down to aggressive spin down or the fact I has desktop drives in a NAS? I have no idea.

I have now moved to a bigger multi core multi drive for my home servers. I now have NAS grade drives in my NAS and set them to defaults when it comes down to spin down. 

Yes this may cost me £10-£20 a year more in costs. But I have had no drive failures (yet) and feel happier overall.

[ThijssjihT]

You sir (assuming you are male, otherwise replace with gender-appropriate alternative), are a hero. Thanks again for sharing your knowledge and experience.

7 hours ago, WilliamsKnights said:

Was this down to aggressive spin down or the fact I has desktop drives in a NAS? I have no idea.

In 2015 I bought my first NAS, and populated with drives I had sitting in my desktop PC. 1 drive failed within a few months, and the other not very much later. After that, I bought NAS drives (WD RED), and they just work, for 8 years now.

I feel like having desktop drives in a NAS is a big reason for drive failure. I will not for a moment consider buying non NAS drives for a NAS after that first experience with desktop drives.

I am not downplaying your spin down argument. I recognize and respect the much bigger amount of experience. Your thoughts on this topic, even if taken with a pinch of salt, are probably more reliable than mine, if I had any on this topic to begin with.

I thought spinning down would be an easy way to get more efficiency from the most power hungry devices in my NAS, and I still want my server to be as efficient as possible, but not at the risk of loosing drives. I will certainly keep the spin-down at default.

Thanks again.

[Dylan]

11 hours ago, WilliamsKnights said:

Was this down to aggressive spin down or the fact I has desktop drives in a NAS? I have no idea.

Great post @WilliamsKnights . I heartily agree that, even with as many data center years experience as I have, there just too many variables to definitively say "this disk".

I have a 10yo consumer grade Netgear ReadyNAS with 2x4TB WD gold drives that are the ORIGINAL drives! They've survived not only a decade but also 4 moves. Now, I've taken really good care of the hardware but I attribute their longevity to both the drive type, the care they've received, but also...just plain old luck.

All of which is to say that, at least as far as hardware goes, do your research, buy the gear designed for your use-case, take care of them and you'll get as much use out of it as you can.

Cheers!