Firstly let me quickly mention that reference to the VNX 7500 48GB and 96GB model is actually 24GB and 48GB per SP respectively.
The Primus needs to be updated as the correct strategy with the large memory models is to allocate the maximum amount possible to Write cache and then allocate the remainder to Read cache. The actual amount will be influenced by certain features (such as FAST, Thin, Compression, FAST Cache) that will take away from the total amount available to allocate to cache.
1) For the 48GB model (or 24GB per SP), the maximum amount of write cache you can allocate per SP is 13450MB
2) For the 96GB model (or 48GB per SP), the maximum amount of write cache you can allocate per SP is 16600MB
For instance, looking back at a VNX7500 with 96GB (or 48GB per SP) that I had been involved with during the initial configuration, the numbers were as follows (per SP):
a) 16600MB Write Cache
b) 7152MB Read Cache
Of course mileage will vary depending on the features that are enabled and the total amount of FAST Cache allocated.
I think you missed a crucial 1 on the read cache though.
my system says 17.552 MB read cache with 16.660 MB write cache
BTW if you want to maximize read cache you can get max 20.152 MB read cache with 14.000 MB write cache.
Note that the read cache figures are per SP and the write cache figures are for both SP (since its mirrored).
So in the last case total cache would be 2x 20152 + 14000 = 54.304 MB
In the first case with max write cache 51.764 MB
I hope you enjoyed your trip to the states!
So possibly we are stating the same thing here, but yes I agree everything is in the perspective of "per SP". So using your array as an example which is as follows:
2) 96GB = 48GB per SP
As you've done (and what is generally considered best practice with the larger memory models) is allocated the maximum amount possible to write cache = 16600MB (per SP) and the remainder to read which for your model = 17552MB (per SP). As already noted, the exact amount remaining is affected by the data services that are active and the amount of FAST Cache.
So doing the math from the perspective of "per SP", we have:
49152MB (48GB per SP)
- 15000MB reserved for system memory (per SP)
= 34152MB available for user to allocate to read and write cache (per SP)
34152MB per SP is allocated as follows:
- 16600MB for write (maximum possible for model)
- 17552MB for read
= 0MB Free memory
array as a whole
Or if we were to look at it from the perspective of the array as a whole:
98304MB (96GB total)
- 30000MB reserved for system memory (total)
= 68304MB available for user to allocate to read and write cache (total)
68304MB is allocated as follows:
- 33200MB for write (total)
- 35104MB for read (total)
= 0MB free memory
I believe we are stating the same thing. However, absolutely if I misrepresented something my apologies. Basically, I simply wanted to note that if you were to use the ~10/90 (R/W) rule, you'd quickly realize that you have memory that you couldn't allocate as you'd bump up against maximums (write cache).
For example, using the allocation above, the ratio is closer to ~50/50 (R/W). As you also point out, there are ways of allocating more to read if you desire, but *generally* speaking as we all know the array benefits from more write cache than read.
Sorry Ganapa, it wasn't until I referenced this post to answer another recent question that I realized there was an unanswered question.
Mirroring is taken into account by the fact that unlike assigning read cache, you only have the single slider for write cache and the amount allocated is identical on both SP's.