This post is more than 5 years old
197 Posts
0
1674
Confirm Data on SSD Strategy is working
I've set a directory to use Data on SSD for its SSD strategy however when running a test workload all the disk activity still occurs on the HDDs. I'm guessing I'm missing a setting somewhere. About this cluster/node pool:
S210 node pool - 4 nodes total
L3 Cache was enabled it is now disabled (same results after disabling)
smartpools is NOT licensed
Directory is set to Manually managed - Data on SSD.
Any thoughts on what could be wrong?
hersh1
197 Posts
0
March 31st, 2015 10:00
Looks like all I needed to do was ask the question in a public forum before I realized what was wrong. I disabled L3 cache under Smartpool settings. I needed to disable it for the nodepool.
Before:
Cluster-1# isi storagepool nodepools view --name=s210_18tb_3.2tb-ssd_256gb
ID: 2
Name: s210_18tb_3.2tb-ssd_256gb
Nodes: 1, 2, 3, 4
Protection Policy: +2:1
Manual: No
L3 Enabled: Yes
L3 Migration Status: l3
After:
Cluster-1# isi storagepool nodepools view --name=s210_18tb_3.2tb-ssd_256gb
ID: 2
Name: s210_18tb_3.2tb-ssd_256gb
Nodes: 1, 2, 3, 4
Protection Policy: +2:1
Manual: No
L3 Enabled: No
L3 Migration Status: storage
kipcranford
125 Posts
1
March 31st, 2015 12:00
You can always spot-check your files to see where they live by doing this:
# cd /ifs/data/testdir
# isi get -DD | less
Scroll down a bit until you see a section like this:
PROTECTION GROUPS
lbn 0: 4+2/2
2,9,190593433600:8192#13
2,9,190594203648:8192#3
2,31,190529372160:8192#16
3,29,190239244288:8192#2
3,29,190238343168:8192#14
3,9,190377672704:8192#16
1,28,190292574208:8192#16
1,29,190211964928:8192#14
1,29,190211686400:8192#2
These "lbn" portions are your file stripes (they're called "protection groups" internally). The syntax is:
, ,
: #Just compare the LNUM from "isi devices" (SSD will be at the top of list, then HDD after) to the LNUM in this output to ensure that the blocks are on the media you think they should be...
kipcranford
125 Posts
2
March 31st, 2015 14:00
It's a sparse region within that protection group / stripe. If that's the only PG for your file, then that likely indicates an area past the end of file. In other words, it represents where something would have been had the PG been completely full. The 0,0,0.. construct doesn't indicate any allocation, it's more just an indicator of a logical construct in that advanced output...
Peter_Sero
1.2K Posts
0
March 31st, 2015 14:00
Thank you! "no allocation" is what I take home here.
Peter_Sero
1.2K Posts
0
March 31st, 2015 14:00
What does 0,0,0:8192#27 mean?
PROTECTION GROUPS
lbn 0: 4+2/2
12,1,355268247552:8192#1
12,1,355268263936:8192#6
12,1,355270082560:8192#4
12,1,370802941952:8192#1
12,1,370803384320:8192#4
12,10,472858992640:8192#16
11,2,477834403840:8192#5
0,0,0:8192#27
13,1,376288796672:8192#10
13,1,376289812480:8192#6
13,17,479887794176:8192#11
13,17,623156789248:8192#5
kipcranford
125 Posts
2
March 31st, 2015 15:00
Correct, no allocation.
As an example, on my 3 node cluster at +2:1 (so stripes of 4+2/2), I create a 480 KiB file, which consists of 3 full stripe units (PGs) and a remaining, partially full stripe unit of 96 KiB (or 12 blocks). The remaining, empty space in that stripe unit, 32 KiB or 4 blocks, is shown in the "isi get -DD" output as a sparse region (0,0,0:8192#4):
PROTECTION GROUPS
lbn 0: 4+2/2
2,10,143425536:8192#16
2,12,276766720:8192#7
2,12,75497472:8192#9
1,31,213786624:8192#16
1,10,176218112:8192#12
0,0,0:8192#4
3,31,110108672:8192#16
3,13,143253504:8192#16
And just to reiterate, any stripe units in OneFS that are not completely full, DO NOT GET ALLOCATED. Only the blocks that have file data in them get allocated...
Peter_Sero
1.2K Posts
0
March 31st, 2015 19:00
Super! Thanks again!
hersh1
197 Posts
0
April 1st, 2015 04:00
Thanks for added info, I remembered that get -DD command from training but couldn't remember what it was telling me. Just tested it out and I can confirm the files are now being stored on the expected drive type.
Thanks!