This post is more than 5 years old
131 Posts
0
1240
Stripe size..
Hi,
Tdev chunk size: 768 KB..
What is stripe size for RAID5 (7+1) TDAT and which way 768 kb distibuted on 7 spindles?
What is stripe size for RAID5 (3+1) TDAT?
What is stripe size for RAID10 TDAT?
which is batter raid to distribute 768 KB for read and write?
I know we need to consider many parameter here but trying to understand flow..
Quincy561
1.3K Posts
0
May 25th, 2011 07:00
That should be a good approximate.
Quincy561
1.3K Posts
1
May 24th, 2011 10:00
All TDEV allocation on TDAT is in 768k chunks, no matter the protection on the backend.
The only real difference in performance is with sequential writes, where 3+1 RAID5 will potentially fill the whole RAID5 stripe allowing for a full optimized write. With 7+1, we will only allocate 768k out of 1,792k (RAID5 7+1 stripe width) and therefore less likely to get an optimized write. Reads or random write performance won't matter if the allocation chunk doesn't match the RAID stripe.
Quincy561
1.3K Posts
1
May 24th, 2011 11:00
Cache is allocated based on device activity and has nothing to with the protection or VP or traditional thick provisioning. A TDEVs have a device WP limit that is the same as a thick device if one was in the same system.
Boom1
131 Posts
0
May 24th, 2011 11:00
Got it..
one more question..
what is cache distribution in terms of read and write per TDEV?
is it depends on # TDEV, # of active path, size?
is there any boundary for read cache per device?
and what is math for write pending track per device?
Boom1
131 Posts
0
May 24th, 2011 12:00
Thank you for answer..
Found following details from white paper..
Symmetrix Enginuity has a logical volume write pending limit to prevent one volume from monopolizing writeable cache. Because each metamember gets a small percentage of cache, a striped meta is likely to offer more writeable cache to the metavolume.
just trying to understand, what is percentage here (4%)? is there any limit for read?
is it same like Max # of Device Write Pending Slots?
I am asking for general device no matter thick or thin..
Quincy561
1.3K Posts
1
May 24th, 2011 13:00
There is no limit for reads. If only one large device is active, it could fill all of usable cache. The most active data should stay in cache because a modified LRU algorithm is used instead of a simple FIFO.
Boom1
131 Posts
0
May 24th, 2011 14:00
I found following details from old thread..
we need 300MB cache for metadata per 1 TB configured storage( DAVE,VDEV,SAVE DAVE) (this is for DMAX not sure about VMAX)
My question is...
if we have 60G cache with 60 TB frame.
I can see available cache solt is almost 48G.
If we consider this frame as a bunker frame and assume we will create 2 VEDV per device. like 20 TB VDEV..( just assumption)
Should i loose 20TB*300MB= 16000MB cache slot?
is it correct math or no?