Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Mute
- Printer Friendly Page
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I'm running a FAS3140 with a 256 GB PAM II card.
This specific 3140 mostly runs VMWare NFS shares and iSCSI LUNs for Exchange 2010.
The Flexcache options with flexscale.lopri_blocks have been turned on and off and below is a sample statistic for each setting.
What I wonder is if the "Disk Reads Replaced" is the most important number to consider.
What are your thoughts?
I put about 100 readings of each into Excel, and found that turning off the lopri_blocks settings gave me a much lower hit percentage, but the "disk reads replaced" average was higher by about 10.
Here are stats for lopri_blocks set to off:
stats show -p flexscale-access -i 4
Cache Reads Writes Disk Reads
Usage Hit Meta Miss Hit Evict Inval Insert Chain Blocks Chain Blocks Replaced
% /s /s /s % /s /s /s /s /s /s /s /s
90 581 66 9643 5 0 1 15 88 199 0 15 88
90 320 48 8852 3 0 6 0 45 124 0 0 45
90 365 58 11284 3 6 1 509 75 173 7 509 75
90 206 70 8846 2 2 1 128 64 139 2 128 64
90 1016 84 12190 7 0 8 0 302 569 0 0 302
90 804 127 12520 6 36 7 1404 155 394 21 1404 155
90 319 75 8569 3 0 6 0 108 210 0 0 108
90 426 141 9220 4 16 6 799 128 269 12 799 128
90 395 50 9931 3 7 1 255 88 228 3 255 88
90 199 45 10401 1 0 4 0 47 118 0 0 47
90 347 61 9870 3 0 0 0 69 152 0 0 69
90 333 113 7005 4 2 1050 127 107 226 1 127 107
90 296 34 8622 3 0 1 0 68 191 0 0 68
90 437 56 11550 3 0 924 15 74 185 0 15 74
Here are stats for lopri_blocks set to on:
Cache Reads Writes Disk Reads
Usage Hit Meta Miss Hit Evict Inval Insert Chain Blocks Chain Blocks Replaced
% /s /s /s % /s /s /s /s /s /s /s /s
94 21 19 11 65 0 0 0 19 21 0 0 19
94 27 22 10 71 0 0 0 25 27 0 0 25
94 147 140 90 62 234 4008 4064 131 145 63 4064 131
94 27 4 9 74 50 3977 2288 10 27 35 2288 10
94 36 12 217 14 0 3 0 15 34 0 0 15
94 22 12 72 23 0 0 0 15 22 0 0 15
94 128 98 116 52 58 56 2976 109 127 46 2976 109
94 11 10 7 60 39 3979 2144 11 11 33 2144 11
94 19 15 33 37 0 0 0 16 19 0 0 16
94 202 193 16 92 0 1 0 190 202 0 0 190
94 154 163 267 36 31 3973 2512 149 152 39 2512 149
94 69 19 87 44 27 0 2272 34 64 35 2272 34
94 89 43 115 43 0 1 0 50 87 0 0 50
94 27 14 11 70 0 0 0 20 27 0 0 20
94 52 49 104 33 93 3990 2560 48 52 40 2560 48
94 29 12 155 15 155 0 2704 15 27 42 2704 15
2 REPLIES 2
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I am not an expert in this subject but the cache hit is more and disk read replaced seems to be more in first scenario, turning on low pri, filer will have to spent extra CPU cycles to handle the additional blocks.
Did you try to take a feedback from end user perspective any difference noticed.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
If you want some more detailed analysis, then the TR's 3801, 3832 and this working paper WP-7107 give a pretty good view of things. You can gain some added performance configurability by using FlexShare (a.k.a 'priority') to prioritize what gets higher caching priority as well. It will also tell you some of the advantages of de-dupe and flash cache/PAM.
Good Luck.
