EMC cache

[Headcrash]

Has anyone any information to describe how the cache is divided up into read and write pending slots.<br><br>Cheers

 

[slimline9]

Assuming you're talking about Symmetrix and not Clariion...
If you get any detail on this please pass it on...however, EMC takes the position that this information is very confidential and never discussed outside of a small group of the development engineers. Given that all reads and writes always pass through the Symm cache, the read/write boundaries will probably change many times. Anyway, nobody outside of these few select, internal, engineers can affect any change in how the cache is used except, of course, to increase the amount of cache...Scott

 

[zuluone]

There are no boundries for read or writes. Cache is divided into slots. Each logical configured device has its on allocated amount of cache slots. This is determined by total amount of cache and tolal number of logical configured devices. Each i/o regardless of read or write takes up a slot. Reads have higher priority till device or frame write pending limits are approched or met. Then writes are giving equal priorty till the write pending lower. If the frame write pendings still go up then the frame goes onto priority destage. This means that all new write i/o are presented a busy til a slot becomes available. (This is a bad thing)

 

[culainn]

just to add to zuluones reply the symmetrix cache does have a boundry for reads and writes strictly speaking. OUt of the host useable cache (not including track id tables and NVd copy etc) the symmetrix will always hold back 20% of slots for reads. This is not to say that only 20% of the slots are used solely for reads that is not true, the cache will be used for reads and writes freely, however if a system write pending ceiling is reached 80% of useable cache the write performance of the array will drop to destage time i..e another write cannot be accepted until a destage operation removes write pending data from cache to disk and thus frees up some slots, the fact that the 20% of cache is serviing reads should mean that read response time is unaffected, however if you are operating at a system write pending limit then effectively your performance goes throught the floor.....system wide..
all the best,.