Practical RAID Decision Making
-
@Dashrender said:
You left out the entire portion of the owners talking to someone else who knows just enough to tell them something else that appears to be cheaper.
Not really, if IT sticks to "that isn't viable" then that's that. No different than if someone was telling the business that they can skip backups. If management is that bad, you have no good options, but IT doesn't have to cave.
-
Tweet, tweet. shared via Twitter
-
@nadnerB Thanks!
-
Nice article, handy information. Thank you.
-
I think you are right on the money with this. I started using RAID 10 single arrays more last year and the payoff has been great for clients.
-
@scottalanmiller said in Practical RAID Decision Making:
The latest on the StorageCraft Blog from me: Practical RAID Decision Making. A whole lot less on the nitty, gritty details and a lot of practical, high level thinking to guide you to quick, simple decision making around spindle-based RAID levels.
I just found this link and discovered that ArcServe bought StorageCraft and removed my writing credits violating my author agreement!
-
One particular situation where I'd find it not quite so straight-forward to go for RAID-10 vs RAID-6 is in a 4-drive setup. Some things to consider would be the performance capabilities of all devices at play (drives, HBA, CPU, etc.) as well as the performance demands of the users/services that need frequent access to said storage. For me, if I/O demand isn't real high for services (and probably using flash, not spindles) I'd be willing to go with RAID-6. Though both RAID levels can sustain 2 drive failures, the caveat with RAID-10 is as long as it's not the same member from each mirrored set. With RAID-6, ANY 2 drives could fail and still be operational and recoverable. I guess it would have to be a very specific concern to opt for the parity overhead in favor of the "added protection" over a statistically very rare potential failure scenario of 4-drive RAID-10.
OK nvm. RAID-10 + backups.
edit
https://www.arcserve.com/blog/practical-raid-decision-making
RAID 10 for four-disk array
Likewise, with a four drive array the only real choice to consider is RAID 10. There is no need for further evaluation. Simply select RAID 10 and continue.Well, SHUT MA MOUTH!
-
@GUIn00b said in Practical RAID Decision Making:
One particular situation where I'd find it not quite so straight-forward to go for RAID-10 vs RAID-6 is in a 4-drive setup. Some things to consider would be the performance capabilities of all devices at play (drives, HBA, CPU, etc.) as well as the performance demands of the users/services that need frequent access to said storage. For me, if I/O demand isn't real high for services (and probably using flash, not spindles) I'd be willing to go with RAID-6. Though both RAID levels can sustain 2 drive failures, the caveat with RAID-10 is as long as it's not the same member from each mirrored set. With RAID-6, ANY 2 drives could fail and still be operational and recoverable. I guess it would have to be a very specific concern to opt for the parity overhead in favor of the "added protection" over a statistically very rare potential failure scenario of 4-drive RAID-10.
OK nvm. RAID-10 + backups.
edit
https://www.arcserve.com/blog/practical-raid-decision-making
RAID 10 for four-disk array
Likewise, with a four drive array the only real choice to consider is RAID 10. There is no need for further evaluation. Simply select RAID 10 and continue.Well, SHUT MA MOUTH!
Consider life expectancy of a RAID 6 over a RAID 10 as well. It's SIGNIFICANTLY more write amplification due to additional parity over RAID 5. RAID 10 would be the best option for having the best possible usage/life expectancy for your drives.
During rebuilds, RAID 6 is the devil. That could be enough writing to make more go belly-up. Then you're toast.
I'd argue that over the lifespan of a server, RAID 10 would likely save more money/resources and headache (and data), making the initial higher cost of capacity worth it. Not only that, but there's other benefits as you mentioned such as speed, iops, etc.
-
@GUIn00b said in Practical RAID Decision Making:
Though both RAID levels can sustain 2 drive failures, the caveat with RAID-10 is as long as it's not the same member from each mirrored set. With RAID-6, ANY 2 drives could fail and still be operational and recoverable. I guess it would have to be a very specific concern to opt for the parity overhead in favor of the "added protection" over a statistically very rare potential failure scenario of 4-drive RAID-10.
Even then, statistically RAID 6 is much more dangerous. RAID 10 has a reliability rating so high that it never matters, RAID 6 does not. RAID 6 has a rebuild time hundreds of times longer than RAID 10; and it has 300% higher URE risks during a rebuild (that is chance of hitting one in a four drive scenario).
Remember the rule of thumb in determining RAID risk: always ignore the false security of "how many drives can you lose." That's not what matters. That's one of many factors, and almost never a significant one, in determining actual risk. URE risks are orders of magnitude more significant and factors like rebuild intensity and rebuild time make "chances to lose another disk" generally more significant than "how many disks can you stand to lose."
At four drives, I know of no scenario where RAID 6 is faster or more reliable than RAID 10. It's always worse. At 5+ drives it starts to have capacity advantages that once in a while make it a good choice. But the rule is at four drives, RAID 6 is a "never" because it's slower and riskier without any offsetting benefits.
-
@GUIn00b said in Practical RAID Decision Making:
I guess it would have to be a very specific concern to opt for the parity overhead in favor of the "added protection" over a statistically very rare potential failure scenario of 4-drive RAID-10.
It's a specific failure scenario that even when it happens, there's no way to know if the same scenario would have been protected under RAID 6 because most scenarios where RAID 10 would fail, RAID 6 would also fail during its recovery mode (nearly 100%.) But the chances that it would face that recovery scenario are higher.
The complexity comes from choosing single unpredictable failure scenarios. After a failure has occurred, if we had the ability to pick how to have protected against it in the past, yes, RAID 6 would be chosen sometimes. There's a known example to explain why you can't use this in real life. It's the seatbelt problem.
Seatbelts save lives. On average, by far, wearing a seatbelt protects you. But there are special cases where the seatbelt can be what causes you to die. Yet statically, you never skip wearing a seatbelt because it is a one in a million chance that the seatbelt will cause a death rather than preventing one. And at the time that you choose to wear or not to sear wear your seatbelt you have no idea which type of accident you will have.
So we know that wearing the seatbelt is the safer bet. Seatbelts are like RAID 10. You can't know how things will go wrong, and in this scenario, RAID 10 protects you much more often than RAID 6 does.
