July 11, 2011 at 1:20 pm
Jim,
Not sure why your wordings were so strong (believe me I was not trying to argue with you). I am here to learn and help/or discuss. I was saying I rememebered that our wins admin configured RAID FIVE with TWO dsik for hp server since the hp server installation provided such option. He showed me and suprised me sicne all I knew previously raid 5 need at least 3 disks. So I said I was not sure how come HP server has option to do so.
Ideally data files should not be put together with the backups therefore in the event of disk failure you might be able to do tlog tail backup to recover from disk failure. In term how to layout the backup stratedy is out of scope of dsicussion. My comment about using raid 10 did not maximize the potntial data loss due to disk failure. Reliable backup is always the key to present the data loss.
Pei
July 11, 2011 at 2:03 pm
Pei Zhu-415513 (7/11/2011)
you can have0 0 1 1
0 1 1 1
(0 indicates failure drive)
This too, is wrong. You may lose any 4 drives out of the 8 as long as no mirrored pair fail. Using your example above the following is possible (or any combination)
0111
1000
1001
0110
1101
0010
Going a step further, remember that with RAID 0+1 you may only sustain a single drive failure in each stripe, so RAID 1+0 as you can see offers better fault tolerance 😉
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
July 11, 2011 at 2:26 pm
Perry,
I undersatnd raid 10 very very well. Previously I just gave one possible situation. Sorry I could not express myself well.
In this case
0011
0011
Data loss will occur.
The example I gave
0 0 1 1
0 1 1 1
would not cause data loss.
Pei
July 11, 2011 at 2:46 pm
Pei Zhu-415513 (7/11/2011)
The example I gave0 0 1 1
0 1 1 1
would not cause data loss.
Pei
please explain why you think this does not cause data loss?
Red are the first mirrored pair, green the second and so on. In your example red shows the failure of a mirrored pair which RAID1+0 cannot sustain. Not even RAID 0+1 can sustain a failure as you have depicted
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
July 11, 2011 at 3:15 pm
Perry,
I was wrong the example mentioned could cause data loss. I was thinking raid 50 I guess. single failed disk can be recovered from mirror. If the whole set of the mirrorred pair failed, it could cause data loss in raid 10.
I apologize to be over confidence.
Pei
July 11, 2011 at 3:43 pm
Pei Zhu-415513 (7/11/2011)
Perry,I was wrong the example mentioned could cause data loss. I was thinking raid 50 I guess. single failed disk can be recovered from mirror. If the whole set of the mirrorred pair failed, it could cause data loss in raid 10.
I apologize to be over confidence.
Pei
Not even RAID50 could sustain that example. RAID60 however could tolerate the failure of any 2 drives in each disk set. The cost of these implementations makes them rare though!
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
July 11, 2011 at 7:35 pm
hi everyone!!!
after all comments/posted in this discussion, i have a question to pei zhu and kevin with relation the proposed scene, what i said in begin of discussion, is wrong?
like pei zhu said, we all have the right to make mistakes, question, propose new ideas and possibilities, because here is a discussion forum, this is what the word means discussion!
July 12, 2011 at 1:39 am
Sizes of Disk and total database size and database growth rate ?
Regards,
Syed Jahanzaib Bin Hassan
BSCS | MCTS | MCITP | OCA | OCP | OCE | SCJP | IBMCDBA
My Blog
www.aureus-salah.com
July 12, 2011 at 6:47 am
rfr.ferrari (7/11/2011)
hi everyone!!!after all comments/posted in this discussion, i have a question to pei zhu and kevin with relation the proposed scene, what i said in begin of discussion, is wrong?
like pei zhu said, we all have the right to make mistakes, question, propose new ideas and possibilities, because here is a discussion forum, this is what the word means discussion!
Everything you need to know about raid10 failures is explained on posts on this forum. I recommend you go to wikipedia or other source for additional reading on RAID topic(s).
Best,
Kevin G. Boles
SQL Server Consultant
SQL MVP 2007-2012
TheSQLGuru on googles mail service
July 12, 2011 at 9:55 am
... Everything you need to know about raid10 failures is explained on posts on this forum. I recommend you go to wikipedia or other source for additional reading on RAID topic(s).
Kevin, with relation to the scene suggested by Jim, on begin of discussion, to four disks use the raid 5 or raid 6 instead of raid 10, where is error?
July 12, 2011 at 10:02 am
RAID5 and RAID6 have the overhead of parity calculations. RAID10 write performance is much better.
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
July 12, 2011 at 10:14 am
Perry Whittle (7/12/2011)
RAID5 and RAID6 have the overhead of parity calculations. RAID10 write performance is much better.
overhead of writes on disks (on case swap disk too), but better performance on reads!
all cases require the detailed study!!
July 12, 2011 at 11:36 am
rfr.ferrari (7/12/2011)
Perry Whittle (7/12/2011)
RAID5 and RAID6 have the overhead of parity calculations. RAID10 write performance is much better.overhead of writes on disks (on case swap disk too), but better performance on reads!
all cases require the detailed study!!
Also, R5/6 is less fault tolerant, more so as you add more disks to the array.
So R5/R6: Faster reads, fewer disks required to implement.
R10: More fault tolerant, faster writes by a big chunk.
So the typical answer of 'it depends' applies here; specifically on if their load will be heavier on reads or writes or both. OLTP systems typically need faster writes. But as stated before, performance is secondary to protecting the data. So fault tolerance makes R10 the winner for me, and the 'default recommendation' without knowing more about the system. And if I can't get the IO throughput out of R10 because of only 2 of the 4 disks being striped, then I'll find a way to add more disks to the R10 array, or tell my customer that they bought too small of a server and to try again. Management can't be feared to the point of where they dream up phantom load requirements and provide cheap hardware. Its the DBA's job to tell them that it is not possible; at least to open up a conversation. I would not be quick to sacrifice data protection by reconfiguring it as R5, putting performance ahead of protecting the data.
Jim Murphy
http://www.sqlwatchmen.com
@SQLMurph
July 12, 2011 at 1:24 pm
RAID10 can offer fast reads depending upon the controller used. Modern RAID10 arrays will read from all online disks to improve performance
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
July 12, 2011 at 1:35 pm
The old Raid 5 reads faster then 10 debate got squashed a bit ago. It's no longer the amazing gain it used to be. This is twofold. First the raid controller software got cleaned up for better performance. Two, the originally mirroring ignored the mirror for simultaneous reading, thus you lost half your drives in performance.
There is still a gain in Raid 5 20 disks vs. Raid 10 20 disks, but it's marginal and you need a rediculous spindle count for it to matter. The primary difference now between the two is cost per gigabyte vs. protection from simultaneous failure. Not all things need perfect recoverability, but many do.
Most SAN controllers (which is what you'd need at this point to see the significant difference) you won't even have direct control of anymore, and merely be exposed a logical LUN for your usage. Best practices or not, you won't even be able to see past the fiber controller. If the manager who can make decisions for both departments is well enough educated in the concerns of database storage, then you might care. Otherwise, you're pretty much left to an email to that team stating "Database I/O disk access is slow, please monitor and correct."
So, since we're dealing with a smaller system that isn't going to have significant data loads until it needs to upgrade to a serious SAN, or at least an external drive array, the best bet here would be to go RAID 10 for safety, since the box isn't going to be regularly monitored for 'red lights', and this will give them the highest margin for error and maintenance.
Never stop learning, even if it hurts. Ego bruises are practically mandatory as you learn unless you've never risked enough to make a mistake.
For better assistance in answering your questions[/url] | Forum Netiquette
For index/tuning help, follow these directions.[/url] |Tally Tables[/url]
Twitter: @AnyWayDBA
Viewing 15 posts - 16 through 30 (of 36 total)
You must be logged in to reply to this topic. Login to reply