HPE Storage Users Group
https://www.3parug.com/

3par SSD failed chunklets
https://www.3parug.com/viewtopic.php?f=18&t=2879		 Page 1 of 1
Author:  michawel [ Fri May 18, 2018 9:10 am ]
Post subject:  3par SSD failed chunklets
Hi, I need to clarify one important thing.
When SSD drives are weared by write operation there is a percentage level (100% to 0%).
OK, but is this percentage of wear dependent on failed chunklets?

I have real case, we have 8 SSD 1920 drives in 7200c and then:

62 3:14:0 SSD normal 1787 902 0 706 0 0 1 0 0 178 0 0
All drives are reported like this.

So 1787 available chunks, 902 chunks used, 706 free, 1 failed and 178 as spare area.
This we have for all 8 drives (datas are well balanced), so we have 8 failed chunks on all drives.
We asked about this to HPE and answer is from my point absolutely unbelievable:
It is by design when every 1% of wear will cost 1 chunklet for this type of drive (there is calculation 1787 x 0,1 = 178,7 = 1,78 of failed chunk)
I can´t believe it, because spare capacity I understand as reserve for failed drive.
I understand that 3PAR is not rebuilding whole disk but only production (used) chunklets, but what in case the array is almost full (we are not in this situation) but lot of customers are (really want to utilize expensive SSD tier).
I thought that 3PAR drives have some reserve (service, hidden) chunklets for this situation. Can somebody confirm or deny this?
Author:  ailean [ Fri May 18, 2018 10:10 am ]
Post subject:  Re: 3par SSD failed chunklets
My understanding is that 3PAR added a feature (via drive firmware) that made all the typical reserved remap space on the SSD available as usable space from day one (increasing the available space for customers).

Failure would be handled at the array level with the wide stripe and reserved spare space for drive failures.

The draw back, as you've seen, is that the SSD wear now eats into this extra usable space as the drives get older, slowly shrinking the array. However the extra space would never have been accessible if they'd not added this feature. ;)
Author:  michawel [ Fri May 18, 2018 10:20 am ]
Post subject:  Re: 3par SSD failed chunklets
Ok, but we have 1787 chunklets on 1,92TB. It doesn´t look as reserved space.
firmware level is 3.2.2 MU6 now, but was 3.2.1 MU5 when this occured.
I am maybe little bit out, but system shows only 1% of wear level (99%).

I am asking because I would like to have 1 disk in reserve for failed drive status and immediate rebuild. According to numbers I can see from array I don´t have enough spare space for totally full drives.
Because 902 + 706 = 1608 chunks and spare space are now 1424.

Thanks for clarification.
Author:  ailean [ Fri May 18, 2018 10:42 am ]
Post subject:  Re: 3par SSD failed chunklets
I suspect the size difference is TiB vs TB.

The sparing ratio is defined when the array is first setup, with small numbers of drives (3PAR was designed with 100s of drives in mind) the default ratio often panics customers (eating up something like 2.5 drives worth of space) so often people select the lower ratio option on small systems.

Also you should never run a thin array to the point that drives are full (70% is a typical healthy level), either buy more space or move/delete data, if I haven't got a purchase order out by time they reach 90% I start panicking. :shock:
Author:  michawel [ Fri May 18, 2018 12:02 pm ]
Post subject:  Re: 3par SSD failed chunklets
thanks

Understand. about capacity yes, as you see we are not full and I warn all users about this. There is second reason for to be at max. 70% max. 80% - for LUN manipulation and also snapshot space etc.. That is all ok.

For less than 24 drives (I think, not sure), spare is set to 1 drive capacity, for more default is as you wrote (two drives).

I am very sensitive about storage so for me this was quite bad suprise as I wrote before. I thought that there is reserved space for weared blocks. Especially due to disk type (more than DWPD 10)

You wrote customers are in panic, yes I saw something similar here in near history. But you now how much cost SSD for 3PAR and lets calculate we have 8x3.84TB SSD (similar price to Mercedes C class after discount) - spare capacity cost 1 drive OK, RAID6 (recommended by HPE for this type) cost -2 drives. So from 8 you have 5. Very cost effective. :) This is just glossary.
Author:  ailean [ Mon May 21, 2018 6:01 am ]
Post subject:  Re: 3par SSD failed chunklets
Yeah I do admit when I read the feature notes about the remap and how arrays would shrink over time I was like EKKK! :o But I've kind of got used to the idea now and I can see some sense in how they are doing it (rather then lots of individual drives doing there own thing when they feel like it and causing unpredictable performance the array simply handles it as it would any bad block).

The RAID6 thing also bothers me but I'm also worried about the drive sizes so they kind of balance out in my head, the performance impact is still an issue however the space side isn't because in 3PAR a RAID5 3:1 actually uses the same amount of parity space as a RAID6 6:2.

However these things are still a little tricky in small systems, I had to setup our first 8000s as RAID10 because there were too few drives to do anything else, luckily with software updates and cheap conversion to single license I've now managed to migrate them to RAID5. More space for free is always good.

But yes 3PAR is not cheap, the minimal 8200 system is a lot cheaper then some kit though but for pure cost we still run two other vendors, although I wouldn't want any of our critical systems on those. ;)
Page 1 of 1 All times are UTC - 5 hours
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/