Unsolved
This post is more than 5 years old
2 Intern
•
308 Posts
2
13497
Understand the “Write Hit” and “Write Miss” in Symmetrix
Understand the “Write Hit” and “Write Miss” in Symmetrix
Introduction
This article describe the differences between the “Write Hit” and “Write Miss” and how they impacts the write IO’s performance in Symmetrix array.
Detailed Information
A Write Hit, also called Write Pending Write which happens when there is sufficient room in cache to store a requested write IO. The host director finds an available cache slot and pipes the host data to it. The host director immediately finishes the operation with the host, signaling that the data was successfully written. The cache slot used for the write is marked with a special “write pending” code to indicate that it cannot be reused until the data is safely stored on disk.
A Write Miss happens when Symmetrix cache slots in which to store writes, run out, a new write request cannot be serviced as fast as a regular write hit. This operation is commonly called a Delayed Fast Write or Write Miss.
Based on above description, the Write Miss requires one more step than Write Hit, which is cache allocation. But this cache allocation process is very quick. It means there is not any performance difference between Write Hit and Write Miss. In an extreme case, the Write Miss has negative performance is once the Writing data reach the system level cache limitation. (75% of available cache) So when a request for a write occurs that cannot be serviced because of lack of cache, existing write pending data is immediately destaged by disk directors. When the host director observes that there is space in cache, the write is completed as normal. This is where the term Delayed Fast Write originates: it is a normal Fast Write that must be delayed while destaging is done.
Following formulas illustrate the time structure of those three Symmetrix write operations:
- Write Hit = Write data the cache slots (64KB tracks) where is allocated and data existed.
- Write Miss = Allocate cache slots + Writing data
- Delayed Write = Destage data into disk + Free up cache slots + Writing Data
As the cache slots allocation is very fast. From performance perspective, usually we don’t care about the “Write Hit” and “Write Miss”. But we have to consider if write IOs are “Delayed Write”. That’s why we should check if the System Write Pending value (75% system cache) is reached. Or if the Write Pending limit of individual device (5% system write pending) is reached.
iEMC APJ
Please click here for for all contents shared by us.
Quincy561
1.3K Posts
0
July 8th, 2015 22:00
Some of this needs some corrections.
A write hit to the tools such as STP and Unisphere is a write to cache that does not have to wait, for example if the system has no write space, such as in a maximum system WP limit, the writes will not be hits, but those are delayed waiting for destage. A miss is when the write has to wait for something like write cache or a slot lock, etc.
The write pending writes are writes to cache slots already allocated and already waiting to destage, but have not yet done so. Overwriting those slots is easier than getting a new slot and making it write pending. This is called a write hit to tools like SymmMerge. The other type of write in SymmMerge is a LRU write, which is where a slot must be allocated. SymmMerge assumes that write hit writes never destage, and LRU writes always destage.