-
Promise Pegasus drives
-
[Brett Sherman] “A lot to be skeptical about.”
If you have a component with an MTBF of five years, a system with two such components working side-by-side will have an MTBF of 2.5 years. When you have racks with hundreds of drives, those MTBF numbers quickly become quite meaningful.
That doesn’t mean MTBF numbers are real though. Google didn’t find them so, for instance. It could be a very nice and real conspiracy and there could be a slew of class action suits brewing. Or, it could be just the reality of the industry where reliability is such a complex issue, that it can’t be reliably (pun intended) measured.
For single drives, AFR numbers are easier to digest though.
-
Statistically, you can’t do that sort of multiplication with MTBF. If it were “Average Time Before Failure” maybe. But MTBF simply means half fail before, half fail after with no indication of the distribution of when they fail. So 50% could fail within 1 minute of plugging in and 50% could fail after 300,000 hours with the same MTBF. Theoretically of course.
I’m not sure it’s a conspiracy. But I definitely think manufacturers are generating the numbers that they want. If manufacturers wanted to provide more better reliability data, they could. DOA rates, failure within 1 month rates, Scan errors per X time, all of those would likely be more predictive of reliability. But the manufactures like MTBF because it indicates a staggered reliability from desktop to enterprise without indicating significant differences within the same class of drives. If it were meaningful every single model should score differently.
-
[Brett Sherman] “Statistically, you can’t do that sort of multiplication with MTBF. If it were “Average Time Before Failure” maybe. But MTBF simply means half fail before, half fail after with no indication of the distribution of when they fail. So 50% could fail within 1 minute of plugging in and 50% could fail after 300,000 hours with the same MTBF. Theoretically of course.”
Agreed, if the distribution of failures over time was unknown. Assuming it’s spread relatively evenly over time, with no correlation between failures on different disks (which is an iffy assumption, I know), can we then do that kind of multiplication?
-
I’m not sure you can derive a whole lot from MTBF at all.
Look at table 2 in this study – https://static.usenix.org/events/fast07/tech/schroeder/schroeder_html/index.html
There is absolutely no correlation between MTTF and real world ARR. It seems that the take home lesson is plan to replace drives whether they are enterprise or desktop. This also reminds me I’d better start replacing my archive drives that are over 5 years old…
-
If MTBF meant something real then we could all use RAID 0.
If I can use desktops (because they are tested and approved by the RAID vendors) then I will for my own in-house RAIDs.
We only used desktops in customer RAIDs because the drive vendor also sanctioned them (Hitachi approved the Deskstars before WD got hold of them – for workstation RAIDs 24/7).
It’s all pretty simple and inarguable.
-
Health monitoring and sufficient protection against UREs and multiple failures is something to rely on, not MTBF. In that, you’re right. Replacing all drives after five years without any other indications of impending failure? I’d have to see a more robust study to sign up for something like that.
(This Schroeder/Gibson paper admits to a rather dramatic lack of consistency in data collection and fault reporting, making it a questionable source for MTTF assessments. Add to that wildly ranging sample sizes, definitions of “fault”, data collection periods, and I am not sure how you could make any conclusions from that paper.)
-
I guess I should say, I should start thinking about replacing all my archive drives over 5 years old. Which for me, means I will get around to replacing them in a couple years. 🙂 And I’ll start with the ones that are 10 years old.
These are drives sitting on the shelf that I don’t spin up much if at all. If they were active, I wouldn’t worry about it much since I would know when they fail and immediately make a new backup. And the costs to move them to new, larger capacity drives is not great in the scheme of things. Not to mention it reduces the number of drives I have to keep around because I can condense 5 500GB drives onto a 3 TB drive.
Since I work for an organization that’s been around for 100 years, I have to think more about archiving for historical sake. Most post houses don’t last long enough to worry about it. 🙂
-
Although the original post is over 2 years ago, I thought I’d share my recent woes with a Promise R6 using the recommended Seagate drives.
Over the last 3 years, we’ve had 7 Seagate drives fail in the two raids, and despite querying this with Promise support, all they can offer is to use the Seagate or Toshiba (discontinued) desktop HD’s.
Tech support say that non of their DAS raids support enterprise or RAID specific (WD RED or RED Pro) drives and that we’re on our own if we change models away from the only two drives they list!So looks like it’s bye bye to Promise kit.
I can only say from past experience of SCSI through to SSD, Seagate have been the worst drives in our history, followed by the cheap WD Green & Blue. Hitachi unto a few years ago were fine, and recently the WD RED (in 4 to 6 bay RAIDS) & RED Pro have been the most reliable.
-
We here at Maxx Digital have had our best Success with HGST Enterprise Drives.The price of drives is at an all time low. A customer can buy 96 TB of storage for less then a 1TB array not that long ago. Not sure why people would compromise the hard work and use anything but Enterprise.Just my two cents. Happy Thanks Giving to all.
Sincerely,
Ron Amborn President
Maxx Entertainment Digital
3189 Redhill Ave #B
Costa Mesa , Ca 92626
Direct 714-374-4944
Cell 714-713-4492 Fax 714-374-3404
ron@maxxdigital.com
http://www.maxxdigital.com -
Ron, 100% agree with you, sadly Promise don’t support enterprise drives in the R6 or any of their DAS RAIDS!
I spoke with their tech support and they say unless we use the Seagate or Toshiba desktop drives they won’t support us.
The real joke is that they can’t be bothered to test any drives that have been released in the last 3 years, so at some point none of their ‘recommended’ drives will be available to purchase, leaving all R series users in the proverbial …
Reply to this Discussion! Login or Sign Up