HDD Characteristics in Terra/Prime servers?

[DieZwei]

I totally get the rationale behind, just trying to put it into perspective.

 

[by96]

I lucked out and got an 8TB Compact Prime used with two years of warranty left (thought the prior owner barely used the server). I have five or six favorites downloaded to keep, but mostly download, watch, and delete movies for myself and my family. I asked before about the lifespan of the SSD and the response (which was very helpful) suggested at minimum, 4 years of a daily download of a 100GB movie, though the lifespan could be twice that too. I probably download two or three movies a week, but many are HD and pretty small (as I am loving finding old movies). I figure that given the normal life of an SDD, I hope to get well over four years, as I don't see myself affording a new SSD server. I could afford an 8TB Compact HDD in the future, but nothing more.

The idea of being able to replace the SSD in my Prime after warranty, even if they wanted $2000 for it, would be so amazing. I get it is a long shot, but it is frustrating to know SSDs and HDDs are replaced all the time in so many devices, but they won't do it for these very expensive servers.

 

[artur9]

... the lifespan of the SSD and the response (which was very helpful) suggested at minimum, 4 years of a daily download of a 100GB movie, ...

I think I posted in another thread, when I asked the Google AI about this it said the SSD would likely outlive the electronics of the device containing it. I asked something along the lines of 2TB weekly download, iirc, and it said it would last 10years (baring the statistics of MTBF etc)

 

[LoveCinema]

This topic has come up a couple of times on the forum and I think people universally support something to avoid consumers left holding bricks.

However, I should point out that previously K had a line-up of the Strato C (no drive), a couple of Compact Terras (single drive), and numerous Full-Size Terras (multiple drives). Therefore, the "potential bricks" were a small portion of their products.

With drives becoming standard in players, with the introduction of the Mini Terra, and the elimination of all but the 120TB Full-Size Terra, K's product line is now nearly entirely built with "brick potential." I sincerely hope that this fact resonates with K and results in future announcements.

I can state that previously customers were very supportive of quite expensive options, meaning that K would have margin - if not bandwidth - to provide offerings.

 

[OCD2]

I think I posted in another thread, when I asked the Google AI about this it said the SSD would likely outlive the electronics of the device containing it. I asked something along the lines of 2TB weekly download, iirc, and it said it would last 10years (baring the statistics of MTBF etc)

I don't believe everything that Google AI (or any AI) says. What really counts is the number of terabytes written to memory over a period of time and the quality of the SSD design. Is AI saying the mean time is 10 years or that it could last 10 years?

 

[RAGJr]

Back in August I was able to purchase a new, unopened Terra Prime 96TB. Since I don’t have a media rack and the media cabinet is full, I decided to place the server on the backside of my desk behind my monitors. In doing so, I’ve been able to learn a bit about its behavior since I’m sitting near it for many hours each day.

• It doesn’t appear that the spindle disks ever spin-down. I hear a near-constant read/write operations going on across the disks and can feel the vibration from the disks spinning at all times. There doesn’t appear to be any sleep or power-saving operation.
• The disks sometimes do a surprisingly large amount of work operations at odd times. I can only assume they’re rebalancing load across them? I’m not entirely sure. This is especially true with recent download activity, which makes sense.
• An observation: the download and write speed of the server is cut by at least half (from 1gpbs to ~400mbps) if a player is being used to watch a movie. Not entirely surprising, but I was surprised it was such a big impact from just a ~100mbps stream.

I’m not complaining about anything, these are just my observations. That said, I would love it if Kaleidescape would add some sleep and power saving operations to the logic of these servers. Doing so would help prolong the lifespan of the disks and conserve a tiny bit of power.

Possible logic:

• If a player within the system isn’t active for X time, or if they’re all in sleep mode themselves, enter sleep mode.
• Perform server maintenance operations as-needed or by remote command.
• Upon player awake, perform an awake of the server.
• Allow an option in the server Settings menu to toggle this functionality on/off.

 

[cyberkryten]

This is all totally expected, to address your three bullets:

I'm sure this has been discussed before on here, but KS will be using Enterprise drives in these rackmount devices and those drives will be rated for 24/7 operation. At a datacenter level, we never power down drives! Spinning them up/down takes time and is the most 'risky' point in a drive's life, so I cannot see why KS would add a power down option.

Disks often 'make noises', this is often NOT anything to do with the data being read/written to the drive but internal operations, head movement or other things that are contained entirely within the firmware of the drive and not commanded by the attached system. This would include bad sector maintenance and data relocation. If KS are doing some sort of RAID, then there are often periods where you scrub the data to confirm the parity is still correct. Again, that's done at the RAID controller level (probably software these days) but still below the level of any intentional 'work' being done by the KS system itself.

Spinning disks are BAD at doing both read and write at the same time. For operational systems, we optimise for read performance, or write performance, you just cannot optimise for doing both sustained read and write operations at the same time - the disk is spinning and only has one read head so performance falls off a cliff (one of the MAJOR, advantages of SSDs is their ability to instantly move between read and write operations with no delays - you don't have to wait until the disk has spun under the read head, which has moved in/out to the right location)

 

[RAGJr]

This is all totally expected, to address your three bullets:

I'm sure this has been discussed before on here, but KS will be using Enterprise drives in these rackmount devices and those drives will be rated for 24/7 operation. At a datacenter level, we never power down drives! Spinning them up/down takes time and is the most 'risky' point in a drive's life, so I cannot see why KS would add a power down option.

Disks often 'make noises', this is often NOT anything to do with the data being read/written to the drive but internal operations, head movement or other things that are contained entirely within the firmware of the drive and not commanded by the attached system. This would include bad sector maintenance and data relocation. If KS are doing some sort of RAID, then there are often periods where you scrub the data to confirm the parity is still correct. Again, that's done at the RAID controller level (probably software these days) but still below the level of any intentional 'work' being done by the KS system itself.

Spinning disks are BAD at doing both read and write at the same time. For operational systems, we optimise for read performance, or write performance, you just cannot optimise for doing both sustained read and write operations at the same time - the disk is spinning and only has one read head so performance falls off a cliff (one of the MAJOR, advantages of SSDs is their ability to instantly move between read and write operations with no delays - you don't have to wait until the disk has spun under the read head, which has moved in/out to the right location)

Yes, and no. Like yourself, I'm also technical and love the nerdy details

While it is true that a constant spin down and up can be bad for any mechanical motor, there is a happy medium to be found where it becomes beneficial to do so, especially considering the spin lifespan limits. Enterprise drives or not, spinning disks will fail, and they fail in data centers all of the time, often at very high rates. In my case, I typically use the Kscape on the weekends. Having it sit and spin for 5 days during the week seems unnecessary. I did add in the idea to have this be an option to have either on or off to allow the user to make their own determination for these and other factors.

And yes, I understand the concept of RW performance. The thing is, I don't know what is actually in the server. Is this a RAID system? If so, it's only RAID 0. Likely software RAID, not hardware, so the enhancements of a dedicated controller with processing and caching are mute or less. If it's just JBOD, are the files written/read across all disks or does an entire file sit across a single disk? I don't know what type of disk-level or file system level load balancing is being done; it's just a curiosity.

I have a bit of experience with this sort of thing having built several Tier 4 data centers, one of the first commercial cloud computing grid platforms which hosted (at one time) about .5% of world wide web namespace on my servers and having gone through probably millions of disks across my hosts

 

[DieZwei]

The idea of putting a K server to sleep when not in use has already come up in the past (surprise ).
I´d also appreciate it. Especially for use-cases outlined above (one or two times use a week), it feels odd to have a server running 24/7. We´re not running a HA environment here.
Might be a German thing, but I´d like to shut down equipment when not in use. A lot of people have their NAS drives spin down when not in use without issues.
However, I understand it´s not an easy task to introduce such a feature if the system (OS, RAID, filesystem) doesn´t support it. But would be great to see it at some point in time.

 

[cyberkryten]

Failure rates for drives that have high spin up/down rates go through he roof at ~ 3 years at the datacenter level and KS offer a 5 year warranty!

In home-use systems, that's higher (based on disk return rate analysis) - likely because of thermal shocks. Homes are not (outside USA!!) temperature controlled in the same way - most people put NAS devices in a cupboard so get huge temp variations. Spinning drives will mostly have a constant temp (or at least varying slightly but the seasonal temp).

It all depends on the exact disks/spec/config used, which we don't really know.

Worth remembering that the only real reason to consider spinning down disks would be a) to reduce noise or b) save power

KS don't live in your house or pay your power bills, and they offer SSD based systems which address both issues, so I suspect there's little to no incentive for them to add functionality which may affect warranty claims

 

[J.Green]

You can always power down your server. Just press and hold the front panel power button for 3 seconds until it shuts down (it's behind the faceplate on the full-sized Terra). Then press the button again when you want it powered up.

NOTE that if you do this, the information in your local movie guide database will be out of date and any downloads you queue -- especially if they are new releases -- will get blocked until an updated movie guide has been downloaded and processed.

 

[DieZwei]

Failure rates for drives that have high spin up/down rates go through he roof at ~ 3 years at the datacenter level and KS offer a 5 year warranty!

Well, I wouldn´t call spinning down and up twice a week as "high spin up/down" rates.
But I agree to

KS don't live in your house or pay your power bills, and they offer SSD based systems which address both issues, so I suspect there's little to no incentive for them to add functionality which may affect warranty claims

 

[torval]

You can always power down your server. Just press and hold the front panel power button for 3 seconds until it shuts down (it's behind the faceplate on the full-sized Terra). Then press the button again when you want it powered up.

This is the way I do it.
I watch a movie every day, but don‘t want to have the server running 24/7.

 

[FURY]

This is the way I do it.
I watch a movie every day, but don‘t want to have the server running 24/7.

K Tech told me not to power down my compact terras, on a daily or even weekly basis, as it can do bad things to the drives, which are intended to be left running. I on several occasions have powrered down the servers and players for weeks at a time as well.

Thank you very much

FURY

 

[torval]

I do the same with my NAS every day since more than 15 years. 2 different NAS and 16 different disks in these 15 years and only 1 disk failure. And These are consumer disks, no Enterprise versions.

 

[Furnace Inferno]

I turned off the disk hibernation on my Synology, when I had it in the house you could hear it firing up the disks many times a day so I just let them spin 24/7 now to stop the repeated spin downs/ups.

Coming up to 6 years old and all still showing as health in RAID10 and 95% full. Once I’ve finished watching my disc backlog though I will power it down to save power and just have the Kaleidescape as the main player and just power it up when I want to use it.