Bring back averaged write option
Posted: Mon Mar 24, 2014 8:35 pm
Hello,
PrimoCache has been excellent, and thank you for the great work.
Would it be possible to add (add back?) an averaged write option? It is my understanding this would have PrimoCache commit to disk in an opportunistic/throttled algorithm while giving priority to read requests and still defering write requests up until either the write cache is full. Once the write cache is full, additional writes are handled in a FIFO cache priority commit.
Currently the behaviour I'm seeing in my workflow I'm hitting an IO bottleneck at regular intervals per the defered latency period which I assume is the write cache commit flushing. For example, setup a r/w cache of 8GB with defer latency of 60 seconds. I burst 3x 1GB writes then read continuously at 60-80% disk load for 60+ seconds. At the defer latency time I witness a full cache write commit and reads stall until the cache commit completes.
An opportunistic cache commit algorithm should queue the writes over the available 20-40% remaining disk load to avoid interval based write cache flush stalls.
For the above example, I'm working on a video editing workstation.
TJ
PrimoCache has been excellent, and thank you for the great work.
Would it be possible to add (add back?) an averaged write option? It is my understanding this would have PrimoCache commit to disk in an opportunistic/throttled algorithm while giving priority to read requests and still defering write requests up until either the write cache is full. Once the write cache is full, additional writes are handled in a FIFO cache priority commit.
Currently the behaviour I'm seeing in my workflow I'm hitting an IO bottleneck at regular intervals per the defered latency period which I assume is the write cache commit flushing. For example, setup a r/w cache of 8GB with defer latency of 60 seconds. I burst 3x 1GB writes then read continuously at 60-80% disk load for 60+ seconds. At the defer latency time I witness a full cache write commit and reads stall until the cache commit completes.
An opportunistic cache commit algorithm should queue the writes over the available 20-40% remaining disk load to avoid interval based write cache flush stalls.
For the above example, I'm working on a video editing workstation.
TJ