HDD Characteristics in Terra/Prime servers?

[RAGJr]

I have some questions on the behavior of spinning HDDs in Terra/Prime servers and was wondering if anyone can provide some insight?

• Is the RPM of the drives used consistent across all systems (i.e. 5400 RPM), or do the Prime variations have higher speeds (i.e. 7200 RPM)? What are the RPMs?

• Do the systems utilize a power saving function to put the disks to sleep and drop the RPMs to zero when not being utilized?
  • If so, is there any sort of noticeable lag upon return to use, or is everything cached to minimize the impact?

Thanks!

 

[josh]

great questions and I've wondered the same. Hope to hear an official answer here soon...

 

[RAGJr]

I would love to know the answers because I'm trying to determine the "risk" of going with a larger Terra Prime server. My experience tells me that spinning disks are a bad idea for longevity. I ran a large data center for many years, and the amount of dead spindle disks was terrifying; literally pallet loads per year. I've not had any device with a spinning disk in it for over a decade, so the thought of spending such an amount on something that will surely experience failures in the first few years of life is not comforting. Albeit, the SSD versions are significantly more expensive, my gut tells me they’ll last at least a decade.

Simple actions like power-saving spin downs not only conserve power, but also extend the life of the HDD.

 

[Rooster]

I’d be curious as well. My general MO has been to try to always be under warranty, given the costs involved.

 

[artur9]

I would love to know the answers because I'm trying to determine the "risk" of going with a larger Terra Prime server. ...

Are those the ones with field replaceable drives? Then the warranty covers you somewhat.

 

[RAGJr]

Yes, but only for 5 years. It may become quite costly after that warranty period is over.

 

[sarahstaar]

This is exactly the concern that's been putting me off a spinning-disk Terra. The point about datacenter HDD failure rates really resonates — that kind of real-world failure volume is a good reminder that "when," not "if," applies to spinning drives. Once a drive's out of warranty, repair options for these servers seem murky at best (which is actually the question I posted about above — would love to know if Kaleidescape/dealers will even touch an out-of-warranty drive swap, and what it costs).


That risk is a big part of why I'm leaning toward the SSD models despite the premium. A Kaleidescape server is a read-heavy device — you write once when you import a movie, then read that file over and over for years. That's about the easiest workload there is on an SSD, since wear is driven mostly by write cycles, not reads or age. So I'd expect the SSD units to hold up well beyond a decade in this use case, while spinning drives carry that mechanical failure risk baked in.


No moving parts also means no bearing wear, no motor start/stop stress — none of the failure modes that make HDDs the weak link in any array. For a device I want to set and forget for 10+ years, that reliability gap feels worth paying for.

 

[gadgetman]

Once a drive's out of warranty, repair options for these servers seem murky at best (which is actually the question I posted about above — would love to know if Kaleidescape/dealers will even touch an out-of-warranty drive swap, and what it costs).
...
No moving parts also means no bearing wear, no motor start/stop stress — none of the failure modes that make HDDs the weak link in any array. For a device I want to set and forget for 10+ years, that reliability gap feels worth paying for.

If a HDD drive fails in an out-of-warranty Terra server with field-replaceable drives, any dealer will simply contact Kaleidescape and order a replacement drive. (Kaleidescape will know which drive failed.) There is nothing "murky" about it. When you receive the new drive, you simply unscrew and remove the existing drive and insert the new drive.

The only question is what size drive the replacement will be, particularly with the storage/memory shortage. So, it's not unlikely that the replacement drive will be larger than the failed drive and therefore more costly. The consolation of paying more is that you gain storage.

Right now, Kaleidescape only offers one SSD option, which is an 8TB Mini-Prime. And, if a drive fails, even though a failure is less likely than a HDD, you have to replace the entire unit at your cost if it is out of warranty. BTW, that 8TB Mini-Prime has two fixed SSD's so, if either fails after the warranty ends, you are out of luck.

So, until Kaleidescape offers field-replaceable SSD drives in a larger Terra server, the conversation is moot. And, if and when they do, the premium cost will probably far exceed the potential cost of replacement HDD drive(s).

 

[DieZwei]

We just had such a case recently (K-1500 failed drive). Enduser cost was roughly €1.750,- excl. VAT (EU is usually more expensive than US) for a 12TB drive. It comes in a cartridge, so no unscrewing/screwing required.
So if the additional speed of SSD is not required, I´d always go for HDDs.

 

[Imakeller]

We just had such a case recently (K-1500 failed drive). Enduser cost was roughly €1.750,- excl. VAT (EU is usually more expensive than US) for a 12TB drive. It comes in a cartridge, so no unscrewing/screwing required.
So if the additional speed of SSD is not required, I´d always go for HDDs.

Sure, IF the HDD has the field-replaceable drives, go for it. Unfortunately, K-scape currently offers too few choices with replaceable drives. Their server lineup needs WORK in my opinion. A lot of people are choosing between the Compact Terra HDD vs the small SSD.

P.S. 2.5gb internet should be standard on all servers and players too.

 

[DieZwei]

Sure, IF the HDD has the field-replaceable drives, go for it. Unfortunately, K-scape currently offers too few choices with replaceable drives. Their server lineup needs WORK in my opinion.

We can easily agree on this for sure.

 

[RobT]

Are there any stats available from K on average life of their installed systems? Coincidentally enough, my neighbour was a software engineer for K when he first graduated and has indicated that the engineering teams work pretty hard on system robustness. I know one day I'll be saying goodbye to my S and its sweet 6 TB, but perhaps that day is far enough in the future that I won't feel too aggrieved when I sprinkle the soil on its dead chassis.

 

[cyberkryten]

You can look at all the stats in the world, but you have to remember that the 'M' in MTBF/MTTF is 'mean' and the 'A' in AFR is 'average' failure rate - these are calculated in a specific way over a huge number of drives (write rates are irrelevant for the usage pattern of downloading movies on KS) and most people see the figures and fundamentally misunderstand them.

MTTF of 1 million hours does not mean your disk is going to last 100+ years, it may last a while, but the longer it runs the more likely failure becomes - most make it to the warrant period, which skews those averages (if few fail for five years that's a lot of catching up to do on the averages). I'd expect to lose about 1% of our drives installed each year, in years 0-5 of 24/7 class enterprise drives used within spec (that means cooled!). After that failure rates go through the roof (most of MTTF is back-loaded pas the warranty end) and so we junk the lot at that point before the replacement burden gets too high.


There's nothing magical about disks in caddies that makes them 'field replaceable', a disk swap is a disk swap whether it took you ten seconds to eject and re-slot a caddy or 2 minutes opening the machine, unscrewing the old disk, disconnect and reconnect cables to the new disk.

However, if that disk was the ONLY disk in the system and it was your boot disk (which in the single disk players, they are) then the issue is that there's nothing to boot up when you put the new (KS approved) disk into the chassis - I suspect that in the multi-disk Terras, the boot code is actually written to multiple (possibly all) drives, so when one fails, the system can still boot, securely and do what is necessary to init a new drive into the security model.

In a single disk KS you'd need some sort of boot device/boot image to start up, initialise the disk and do the security/encryption tasks that sign the disk and tie it to the player. I can absolutely see why KS don't want an unencrypted boot image out there, so that is probably the reason why we won't see 'field replaceable' drives in the single disk players anytime soon.

 

[Imakeller]

In a single disk KS you'd need some sort of boot device/boot image to start up, initialise the disk and do the security/encryption tasks that sign the disk and tie it to the player. I can absolutely see why KS don't want an unencrypted boot image out there, so that is probably the reason why we won't see 'field replaceable' drives in the single disk players anytime soon.

I’d be happy to have an option to send the server to K-scape or a dealer for service.

Also, a 5-year warranty is very good. I wish we could purchase an extended warranty.

 

[RobT]

You can look at all the stats in the world, but you have to remember that the 'M' in MTBF/MTTF is 'mean' and the 'A' in AFR is 'average' failure rate - these are calculated in a specific way over a huge number of drives (write rates are irrelevant for the usage pattern of downloading movies on KS) and most people see the figures and fundamentally misunderstand them.

Seems I'm well into borrowed time by this standard...

 

[Hugo]

My Terra prime 120 is on the way and I was wondering if I can pull 3 of the 4 HDD out of the chassis, and installing HDD’s back one by one into the chassis as my movie collection grows.

 

[HTRaleigh]

Ironically the only computer drive that has failed in any of my computers in the last 10 years was a 3 year old SSD. No warning, it just failed and took my files with it. No normal means to pull data worked. Needless to say, I’m not as enamored with SSD as I once was.

 

[OCD2]

Normally a disk warranty is timed to expire just before the disk is expected to fail. How is it possible that MTBF or terabytes written is only 5 years on high end hard drives and SSDs sold by Kscape which in other uses seem to have life expectancies much longer than 5-10 years especially when used for long term storage rather than frequent re-writes? I can understand an SSD in a standalone Strato V, E, M or K with constant rewrites dying but why long term Terra servers with SSDs?

 

[DieZwei]

Well, I believe most (all?) enterprise harddrives/SDDs have a manufacturer warranty of max. 5 years, so I´d say it would be pretty risky for a company to take responsibilty for additional warranty beyond that for a whole system on top of that.
Seriously, I dont´t think apart from power amps and speakers, there aren´t many devices available with a warranty beyond 5 years.
I´d also really welcome payable repair beyond warranty, but extending warranty beyond 5 years is a bit too unrealistic to ask IMHO.

 

[OCD2]

The Samsung 850 Pro Series SSDs have a 10 year warranty.

I'm simply asking K to be creative and come up with some alternatives.

Let's face it, customers are looking for a reasonable financial alternative to purchasing a brand new server at full profit margin to K. Perhaps the ability to swap with a K approved SSD when errors start appearing or at 5 years with a new replacement that doesn't carry a 100% markup but instead some reasonable discount (perhaps a 30% markup). Perhaps this would be limited to a one time swap. Or when installing the new drive the software recognizes the old drive and wipes it clean before activating the new hardware/ I don't believe this is an unreasonable request. What data centers replace all their servers every 5 years? All computers can be upgraded with new drives, etc. Why not K servers?

I realize this is how K currently generates revenue but it seems rather short-sighted and perhaps not a sustainable business model. Why not make it more palatable for customers by offering some creative solutions for maintenance and sustainability? Anything would be better than the current plan which is essentially nothing = planned obsolescence.