This post is more than 5 years old
80 Posts
1
5947
Symmetrix Mirror Positions
Hi All ,
Naive question ,need some help understanding these mirrors , I've read multiple docs and blogs online and on ECN but the mirror positions ,concept keeps bouncing off my head .I know that prior to 5874 Raid 1/5 occupied 2 (M1,M2)positions each and Raid 6 - one position . 5874 onwards all occupied one mirror position each . That's all I know .
Host I/O > Cache>Mirror>Tdev ? Is this how it works , is the mirror just a temp cache sort of device until the data is destaged to the actual tdevs ?
In simple terms , whats a mirror , why is it there , why won't the data go to the disks directly from cache ?
I would like to know how these work , an example prior 5874 and after will be great .
Edit :(Please attach : Timefinder clone/snap docs , my access is restricted and I can't get any )
Kind Regards
Message was edited by: asceticenergy
rawstorage
419 Posts
2
December 11th, 2014 04:00
Mirror positions were important back in the day, pre 5874 and Pre TF Clone
Every Device had 4 mirror postions _ _ _ _. When a device was configured 2 way Mirror we would consume at least 2 mirror postions so the mirror set would look like m m _ _ . This leaves us with 2 free mirror positions. If we add SRDF this would also consume a Mirror m m F _ leaving one mirror position. This remaining mirror position could be used for another copy purpose. e.g. a BCV device would temporarily occupy this space during an establish m m F B, we could only have a single BCV sync at a time when this was the case, so we'd make one copy then another, if we had 2 free positions we could take two copies at a time. This architecture served it's purpose for many years but there were limitations as you can imagine, when we introduced Clone back in 5671 things changed.
Instead of occupying a mirror position for the clone we use an emulation internally and do everything through pointers, Freeing up Mirror positions that were once consumed during copies, this helped us scale the number for concurrent copies we could take for clones etc.
In 5874 code we changed things again, we have a common code path for all Raid Protections (known as Raid Virtual Architecture)so we no longer consume different mirror counts for different RAID protections we just consume a single e.g. RAID 5 looks like 5 _ _ _, RAID 6 looks like 6 _ _ _ We still have the additional mirrors that can be used for SRDF but local protection doesn't consume any. In summary they are still used internally but they are not as important to the end users as things have changed.
There are a few good KB articles on this, but hopefully you get the idea.
asceticenergy
80 Posts
0
November 26th, 2014 02:00
there are hundreds of posts about this on ecn , I've read everything . Can't get my head around it as yet , hence the re-post . Sorry .
asceticenergy
80 Posts
0
November 26th, 2014 23:00
can someone please explain.
Ugin
122 Posts
2
December 14th, 2014 22:00
I agree with Paul,
Mirror positions is to highlight your devices protection levels. Prior to V-MAX, most devices occupied 2 mirror positions initially unless they were standalone devices, leaving 2 mirror positions available for Local and Remote replication options as there is a maximum of 4 mirror positions.
In V-MAX we see RVA (Raid Virtual Architecture) whereby only 1 mirror position is initially occupied by the device, regardless of whether it is Mirrored,RAID5,RAID6 , freeing up an additional mirror position for local and remote replication options. Hope this explains.
asceticenergy
80 Posts
0
December 18th, 2014 22:00
Thanks guys , cheers !
gautam_sunil
7 Posts
0
December 22nd, 2014 23:00
Hi Paul
Thanks for the clear explanation !!
But I have doubt on Write Operation in case of DMX4 running on 5773 . As per my understanding mirror position is nothing but just a pointer to actual data .
I need some clarification about the Operations on Mirror position . In case of 2 Way Mir device(which is nothing but a RAID-1 configured device ) it will occupy 2 Mirror positions M1 and M2 . So M1 will point to the data residing on Disk1 and M2 will point to the data residing on Disk 2 . And Data will be written as FA->Cache -> DA->M1 ->M2
But in case of RAID5 device it will occupy the 2 mirror position ( M1 and M2 which are mentioned as RAID5, RAID5 ) .
So in this case to which data M1 and M2 will points .
lalithvensri
1 Message
0
September 7th, 2019 11:00
gautham you ave posted excellent question
i have the same doubt since many years. like why this mirror positions are required ( some place it is written data structures etc ) as you said , if raid 2 it is 2 mirror positions are used then only 2 are used for raid 5 and raid 6
even for n number of srdf sessions only 1 mirror position used . lot of questions.
No body answer these questions . except telling some nonsense which every one know.
I did call and ask many times even to EMC support , even they do not have ans.
emc engineering must be having the exact answer , but i could not reach them out.