Jump to content

OT: Any reason not to go RAID 10?


SDS

Recommended Posts

Just now, WhoTom said:

Seems like RAID 5 should suffice - data striping for performance and parity for data security. Is the mirroring of RAID 10 worth the extra cost?

 

 


Retail is $19/drive per month 

Link to comment
Share on other sites

17 minutes ago, Mr Info said:

Would I be correct to assume that AWS is too pricy? RAID Is not really a consideration there.

 

I think so.

 

I'm curious if RAID 10 makes sense in the era of SSDs.

 

I can double the RAM and double the processor benchmark for the price of two SSDs.

Link to comment
Share on other sites

49 minutes ago, SDS said:

 

I think so.

 

I'm curious if RAID 10 makes sense in the era of SSDs.

 

I can double the RAM and double the processor benchmark for the price of two SSDs.

If you have SSDs, you should probably go see a doctor, get some penicillin, or something...

  • Like (+1) 1
  • Sad 1
  • Haha (+1) 1
Link to comment
Share on other sites

We ran that play in High School: it was a fake pitch and quick short pass to our Center who pulled left, and then he threw it back to the QB streaking (yeah, buck-naked) down the right side of the field, while everyone else ran to the left and yelling “Raid 10, Raid 10”  .... yeah, the Bills should put that in the play book.

 

  • Haha (+1) 2
Link to comment
Share on other sites

24 minutes ago, 1ManRaid said:

I came in thinking RAID 10 was some gimmick lineup for offensive tackles...

 

It's a spread formation, 2 TEs and 2 WRs, with each pair of receivers running mirrored routes. In the playbook, it's known as "Data-stripe full-redundancy."

 

 

  • Haha (+1) 1
Link to comment
Share on other sites

15 minutes ago, WhoTom said:

 

It's a spread formation, 2 TEs and 2 WRs, with each pair of receivers running mirrored routes. In the playbook, it's known as "Data-stripe full-redundancy."

 

 

That's a mouthful in the huddle with a loud crowd.  "What was that?  Dates are ripe for transparency?"

Link to comment
Share on other sites

1 hour ago, CSBill said:

We ran that play in High School: it was a fake pitch and quick short pass to our Center who pulled left, and then he threw it back to the QB streaking (yeah, buck-naked) down the right side of the field, while everyone else ran to the left and yelling “Raid 10, Raid 10”  .... yeah, the Bills should put that in the play book.

 

You must have went to St. Joes. I remembered trying to defend that play. But your QB looked more like a Raid 3.

Link to comment
Share on other sites

5 hours ago, SDS said:

 

I think so.

 

I'm curious if RAID 10 makes sense in the era of SSDs.

 

I can double the RAM and double the processor benchmark for the price of two SSDs.

I looked into ssd's vs hdd's a little bit for my enterprise db servers and the drives are designed to last a number of write operations which makes redundancy useless because you would have two drives fail instead of just one. It's just tough for folks like me who have relied on redundancy with spinning disks to let go of that standard, especially because it makes us responsible if there is a physical failure that redundancy would have saved. I haven't seen numbers for the reliability of the ssd's in an enterprise class server vs hdd's, it just seems like the price/value make the ssd's the goto drive at this point. Nvme's are especially fast.?

 

For this site, I would lean toward the ram and processor speed giving your users a much larger bang for your buck. Hopefully someone would monitor the drives for errors and be able to address an issue before it becomes a 5 alarm fire.

 

The other side is how much of a pain would it be to recover from an ssd failure for you. Might be worth having the ability to run with a bad drive and not have to deal with any recovery even if failure is a long shot.

Link to comment
Share on other sites

I'd run RAID-5 on 4 drives, currently running RAID-6 on 8 drives (2 can degrade before data loss). But I run a dedicated rack box with an Areca card. Probably an anachronism by now but eh.

 

Cloud is pretty spiffy now. Hard for the tin cans to keep up. Speed isn't my #1 so it's fine for me.

 

IIRC the hardware RAID makes a big diff over software RAID for speed. Raid 10 would be somewhat equivalent to RAID-6 storage-wise but unless your hardware is super for the RAID-6 (or 5), 10 will do better for performance, as you're not throwing checksums and parity sectors out there. It's just using brute mirroring for your redundancy. A slight bit more susceptible to failure but still an infinitesimal likelihood, and doesn't rely on calculations.

Edited by Ralonzo
Link to comment
Share on other sites

8 hours ago, WhoTom said:

Seems like RAID 5 should suffice - data striping for performance and parity for data security. Is the mirroring of RAID 10 worth the extra cost?

 

 

I thought 10 was combo striping / mirroring. 0 is striping storage across 2+ devices to create a (virtual) single, larger device. 1 is straight-up mirror. 10 is typically 4 drives: 2 x 2-drive stripes. If you had 4 2TB drives a RAID-10 would be a 4TB virtual drive with 1 full virtual mirror. The stripes give performance, the mirror gives redundancy without significant performance degradation.

 

RAID-5 would sacrifice 1 drive for parity block stripes. Each drive will donate 1/n sectors which essentially back up the other n-1 drives in the array by facilitating data reconstruction when data is in an unrecoverable state. Aforementioned 4 x 2TB rack would yield a single virtual 6TB drive (yay) but the tradeoff is performance to maintain the parity blocks across the array.

 

RAID-6, same as above but you sacrifice 2 drives, each drive donates 2/n sectors and your net virtual storage is (n-2)*drive size. The payoff here is arrays with many drives (8 seems popular). An 8 drive RAID-10 will get you 4 of 8 drives worth of effective storage, but a RAID-6 would give you 6 of 8.

 

10 will give very good performance and redundancy on 4-drive configs without fancy hardware. For enterprise use, 5 and 6 may not be as good for performance, particularly with spinners, unless you've a bad-ass dedicated RAID card with copious caching.

 

I won't go into 50 or 60 as that's likely overkill here.

 

  

8 hours ago, SDS said:


Retail is $19/drive per month 

 

Just noticed this... do you get 1 drive gratis? I don't see how RAID-10 is possible with 3 drives ($57/month = $19 * 3, is the inference)

Edited by Ralonzo
Link to comment
Share on other sites

1 hour ago, 4BillsintheBurgh said:

I looked into ssd's vs hdd's a little bit for my enterprise db servers and the drives are designed to last a number of write operations which makes redundancy useless because you would have two drives fail instead of just one. It's just tough for folks like me who have relied on redundancy with spinning disks to let go of that standard, especially because it makes us responsible if there is a physical failure that redundancy would have saved. I haven't seen numbers for the reliability of the ssd's in an enterprise class server vs hdd's, it just seems like the price/value make the ssd's the goto drive at this point. Nvme's are especially fast.?

 

For this site, I would lean toward the ram and processor speed giving your users a much larger bang for your buck. Hopefully someone would monitor the drives for errors and be able to address an issue before it becomes a 5 alarm fire.

 

The other side is how much of a pain would it be to recover from an ssd failure for you. Might be worth having the ability to run with a bad drive and not have to deal with any recovery even if failure is a long shot.

What he said. :w00t:

29 minutes ago, Ralonzo said:

 

I thought 10 was combo striping / mirroring. 0 is striping storage across 2+ devices to create a (virtual) single, larger device. 1 is straight-up mirror. 10 is typically 4 drives: 2 x 2-drive stripes. If you had 4 2TB drives a RAID-10 would be a 4TB virtual drive with 1 full virtual mirror. The stripes give performance, the mirror gives redundancy without significant performance degradation.

 

RAID-5 would sacrifice 1 drive for parity block stripes. Each drive will donate 1/n sectors which essentially back up the other n-1 drives in the array by facilitating data reconstruction when data is in an unrecoverable state. Aforementioned 4 x 2TB rack would yield a single virtual 6TB drive (yay) but the tradeoff is performance to maintain the parity blocks across the array.

 

RAID-6, same as above but you sacrifice 2 drives, each drive donates 2/n sectors and your net virtual storage is (n-2)*drive size. The payoff here is arrays with many drives (8 seems popular). An 8 drive RAID-10 will get you 4 of 8 drives worth of effective storage, but a RAID-6 would give you 6 of 8.

 

10 will give very good performance and redundancy on 4-drive configs without fancy hardware. For enterprise use, 5 and 6 may not be as good for performance, particularly with spinners, unless you've a bad-ass dedicated RAID card with copious caching.

 

I won't go into 50 or 60 as that's likely overkill here.

 

  

 

Just noticed this... do you get 1 drive gratis? I don't see how RAID-10 is possible with 3 drives ($57/month = $19 * 3, is the inference)

I’m glad we have you on OUR team! :)

  • Haha (+1) 2
Link to comment
Share on other sites

14 hours ago, SDS said:

 

I think so.

 

I'm curious if RAID 10 makes sense in the era of SSDs.

 

I can double the RAM and double the processor benchmark for the price of two SSDs.

 

From all accounts RAID doesn't matter for speed with SSDs .  The improvement is barely noticeable - I foolishly went that route RAID 0 with a work machine ~6 years ago - it was pointless (and expensive for the dedicated card).  RAID 10 to make a bigger, more secure against failure disk is all good if that's the aim.

Edited by BobChalmers
  • Like (+1) 1
Link to comment
Share on other sites

10 hours ago, 4BillsintheBurgh said:

I looked into ssd's vs hdd's a little bit for my enterprise db servers and the drives are designed to last a number of write operations which makes redundancy useless because you would have two drives fail instead of just one. It's just tough for folks like me who have relied on redundancy with spinning disks to let go of that standard, especially because it makes us responsible if there is a physical failure that redundancy would have saved. I haven't seen numbers for the reliability of the ssd's in an enterprise class server vs hdd's, it just seems like the price/value make the ssd's the goto drive at this point. Nvme's are especially fast.?

 

For this site, I would lean toward the ram and processor speed giving your users a much larger bang for your buck. Hopefully someone would monitor the drives for errors and be able to address an issue before it becomes a 5 alarm fire.

 

The other side is how much of a pain would it be to recover from an ssd failure for you. Might be worth having the ability to run with a bad drive and not have to deal with any recovery even if failure is a long shot.


right now I can’t think of a reason not to go with NVMe. In fact, I don’t even know why there’s a choice. They are the same price.

  • Like (+1) 1
Link to comment
Share on other sites

I thought from the title that this thread was about Daboll implementing an air raid style offense into his bag of tricks.  Every poster needs to look at a thread like this to make themselves feel dumb and then look at a thread arguing about flats or drums to make themselves feel smart.

Link to comment
Share on other sites

×
×
  • Create New...