Cruising is finally coming into this century
-
@scottalanmiller said:
So here is the RAID 5 question.... Are you saying that I'm wrong and that you've been screwing people over by holding back info that we don't have? Or that I'm right and that you call it narcissistic to be right and stick to your guns while people who know they are wrong try to brow beat others into accepted the wrong things that they say?
No, it's because you have become a serious [moderated] about it. It's not what you say, it's how you say it. Like "Not letting people get away with being idiots isn't the same as needing to be right", but it's actually both. You have to be right, come hell or high water, and you have to stop people from being idiots in your perception. So your line about conceding the argument? Once again, you can claim you are right, bask in that righteousness, and feel better about it. This is exhibited in other topics as well.
You can be wrong, dead wrong. And it's not possible to shape the world the way you want it. If people want to use RAID5, like we do on thousands of servers and SANs, let them. If they want to buy a SAN, let them. Don't expect everyone to bend to your will at any time just because you said so, using whatever "facts" you bring forth. Because some of it is fact, some of it is "fact". You are not that convincing.
-
@scottalanmiller said:
@PSX_Defector for three years I've asked for people to provide math on when RAID 5 would be acceptable. That's a lot if vetting. Not letting people get away with being idiots isn't the same as needing to be right.
But let's face it, once name calling is needed, the argument was already conceded.
I'm not a math person, admittedly. Number crunching has never been my thing. As far as RAID 5 usage goes, any place where fault tolerance would be a benefit, as well as pooled drive capacity would be a good fit. For example, backup targets or DAG member databases are great uses for it.
Mission-critical or high-performance workloads would not be good choices.
-
@PSX_Defector said:
@scottalanmiller said:
So here is the RAID 5 question.... Are you saying that I'm wrong and that you've been screwing people over by holding back info that we don't have? Or that I'm right and that you call it narcissistic to be right and stick to your guns while people who know they are wrong try to brow beat others into accepted the wrong things that they say?
No, it's because you have become a serious [moderated] about it. It's not what you say, it's how you say it. Like "Not letting people get away with being idiots isn't the same as needing to be right", but it's actually both. You have to be right, come hell or high water, and you have to stop people from being idiots in your perception. So your line about conceding the argument? Once again, you can claim you are right, bask in that righteousness, and feel better about it. This is exhibited in other topics as well.
You can be wrong, dead wrong. And it's not possible to shape the world the way you want it. If people want to use RAID5, like we do on thousands of servers and SANs, let them. If they want to buy a SAN, let them. Don't expect everyone to bend to your will at any time just because you said so, using whatever "facts" you bring forth. Because some of it is fact, some of it is "fact". You are not that convincing.
So the point is... stop giving advice and helping people because people are idiots and can get away with not doing the best that they can so, since people don't care and aren't logical then it's best to stop helping them and just allow them to hurt themselves and others.
My theory is... give a man the rope. If he climbs out of the hole, you did the right thing. If he hangs himself, he did the wrong thing. Helping people is never the wrong things in my book. I'd rather piss off ten people who don't care about their jobs or professions if I don't miss helping one person who genuinely is trying to learn or do the right thing. I'm not willing to sacrifice the innocent for the guilty, but will happily allow the guilty to sacrifice themselves if that helps to save the innocent. No one forces people to make bad decisions, that is entirely on them.
-
@alexntg said:
@scottalanmiller said:
@PSX_Defector for three years I've asked for people to provide math on when RAID 5 would be acceptable. That's a lot if vetting. Not letting people get away with being idiots isn't the same as needing to be right.
But let's face it, once name calling is needed, the argument was already conceded.
I'm not a math person, admittedly. Number crunching has never been my thing. As far as RAID 5 usage goes, any place where fault tolerance would be a benefit, as well as pooled drive capacity would be a good fit. For example, backup targets or DAG member databases are great uses for it.
Mission-critical or high-performance workloads would not be good choices.
RAID 5 doesn't fit in those cases, that's the problem. "Fault Tolerance" has become a marketing term that people through around to hide behind. Fault tolerance is a horrible thing if it makes the system more fragile. Like the houses in a hurricane example. Two straw houses versus one brick house. One is redundant or "fault tolerant" and the other is reliable. Sure, two brick houses would be even better still, but that's not the point. Between the two available choices, the redundancy actually introduces the fragility by forces us to increase the failure rate to begin with.
This is why RAID 5 is always bad, it increases the risk out of the gate and effectively does nothing to fix it. The fault tolerant aspects of it are a farce in situations where it is cheap and the cost of making the fault tolerance work makes it too expensive. So RAID 5 either is a bad choice because it lacks the protection that people imagine or it cost to much and is just flaunting that you don't care about spending wisely. Both bad use cases.
-
@scottalanmiller said:
@alexntg said:
@scottalanmiller said:
@PSX_Defector for three years I've asked for people to provide math on when RAID 5 would be acceptable. That's a lot if vetting. Not letting people get away with being idiots isn't the same as needing to be right.
But let's face it, once name calling is needed, the argument was already conceded.
I'm not a math person, admittedly. Number crunching has never been my thing. As far as RAID 5 usage goes, any place where fault tolerance would be a benefit, as well as pooled drive capacity would be a good fit. For example, backup targets or DAG member databases are great uses for it.
Mission-critical or high-performance workloads would not be good choices.
RAID 5 doesn't fit in those cases, that's the problem. "Fault Tolerance" has become a marketing term that people through around to hide behind. Fault tolerance is a horrible thing if it makes the system more fragile. Like the houses in a hurricane example. Two straw houses versus one brick house. One is redundant or "fault tolerant" and the other is reliable. Sure, two brick houses would be even better still, but that's not the point. Between the two available choices, the redundancy actually introduces the fragility by forces us to increase the failure rate to begin with.
This is why RAID 5 is always bad, it increases the risk out of the gate and effectively does nothing to fix it. The fault tolerant aspects of it are a farce in situations where it is cheap and the cost of making the fault tolerance work makes it too expensive. So RAID 5 either is a bad choice because it lacks the protection that people imagine or it cost to much and is just flaunting that you don't care about spending wisely. Both bad use cases.
How would RAID 5 be riskier than RAID 0, which would be the other high-capacity option? With 0, if you lose a drive, you lose the array. With 5, if you lose a drive, there's a chance you might lose the array.
-
@alexntg said:
How would RAID 5 be riskier than RAID 0, which would be the other high-capacity option? With 0, if you lose a drive, you lose the array. With 5, if you lose a drive, there's a chance you might lose the array.
Because in RAID 5 you are more likely to lose a drive than in RAID 0. So the risk to the array experiencing an "event" is higher in RAID 5 (assuming the same or better speed and capacity needs are met.) Both are essentially unprotected. Using the might survive a drive loss is extremely misleading. In a modern array that is like jumping out of an airplane - yes with good training and if you keep your head and you are not over water you might land and not die, but the chances are incredibly low.
With large, modern drives, in an array of any size (typically 5+ drives) the extra risk of array event introduced by RAID 5 equals or exceeds the might recover chances making RAID 0 meet or beat the reliability of RAID 5 while being faster and cheaper at the same capacity (faster if there is any write load, RAID 5 is marginally faster in most pure read only scenarios, but if read only we typically don't care about reliability anyway.) The larger the array, the safer RAID 0 becomes in comparison to RAID 5.
Of course, never is either safe. Both are ridiculously dangerous. But people confuse the might survive a disk loss event with being safe, which it is not.
The only ways to make large arrays (large meaning 1TB+) safe is either through escalating levels of parity (RAID 6, then RAID 7.3, theoretically RAID 7.4, etc.) or by jumping to redundancy (RAID 10.) There is no means of protecting large arrays without spending more to do so in terms of sacrificed capacity.
-
The RAID 0 to RAID 5 risk ratio changes over time. For the first few days, RAID 0 has a huge advantage. Then for the next 24 months it loses a little, then it steady gains over time. It depends on the individual array configuration both in terms of drive sizes, drive reliability, URE and total array size taken together to determine at any given moment which type would be more reliable. For smaller arrays (especially small drive sizes, small over all capacity but many spindles) RAID 5 can certainly pull ahead in terms of reliability, but it never gets far enough ahead to be taken seriously because "marginally safer than RAID 0" is still a major reliability problem. But for larger arrays, RAID 0 starts safer and remains so over the entire lifespan. It varies, but both are just like setting your data on fire. Once data is deemed so worthless, stateless or volatile that no protection is needed, then RAID 0 is fine.
-
Here is a quick comparison of the non-reliability factors of two arrays, assuming 75/25 read/write IOPS blend. The two arrays have the same capacity of 22.3TB.
RAID 0 IOPS: 589
RAID 5 IOPS: 379RAID 0 Cost: $1,640
RAID 5 Cost: $1,845Even leaning heavily to reads, the RAID 0 saves us money and delivers more performance. So the non-reliability advantages are clear. Once you do the math and see that RAID 0 is actually safer than RAID 5 in this array too, it is a win/win/win. Now, like I said, both are horrible choices, but because RAID 0 is faster, cheaper and safer in modern sized arrays, we can eliminate RAID 5 from consideration for these kinds of scenarios right off of the bat as it offers no upsides at the array layer.
-
Another way to think of it is with cars. You are sending someone to the store. You need a lot of stuff and you will make these trips every day. You have the choice of two cars, one is a tiny bit smaller but has extra safety features that make it slightly safer than the alternative car that has just a little more storage capacity.
The result is that one car needs to go to the store, every day, eight times to get everything that is needed. The other car has to make the trip nine times, while being marginally safer each time. Let's say that car A actually is pretty unsafe, so unsafe that any accident means certain death to the driver. The other car has a 1% chance of saving the occupant. So 99% of the time, they will be dead too.
Now let's say that there is a 2% chance of having an accident on every trip to the store. Which is safer?
Clearly, the car with zero safety but larger capacity is safer... why? Because it takes fewer chances on trips to the store. Each trip is incredibly risky and the protection in the safer car doesn't come anywhere close to off setting the risk from that extra trip to the store each day.
But we can also see, that both scenarios are ridiculously dangerous, but the car with no "safety" is safer than the one with "practically no safety."
This is incredibly close to how RAID 0 and RAID 5 work. RAID 5 needs more disks and pushes them harder making the chance of an event happening higher. Higher than its parity can protect against.
-
See, this is what I'm talking about. You are being a [moderated] posting hundreds of times for RAID5 because "reasons". And you have to be "right" by posting hundreds of posts backing up your [moderated].
