Place a limit on deferred Blocks Topic is solved

[rit]

With the release of 1.0 I have decided to try the cache on my live system.

One thing that is clear is that in my environment at least it would be good to be able to limit the total number of outstanding deferred blocks. I've defer-writes set to 1s (would be nice to support .5s) as all I wish to do is catch the re-writting of blocks caused by database work, but as I also use virtual machines on my system the number of deferred blocks can hit 100,000+ when I suspend a VM and a long wait for these blocks to be written back to the underlying disk(s).

It would be helpful if I could set a limit on the number of deferred blocks that the program will allow before it just slows down its acceptance of new blocks based on the speed at which it is able to write out blocks to the underlying disk(s).

My only solution at the moment is to turn deferred blocks off, which is a shame.

[InquiringMind]

If you have that much data being written, then Primo Ramdisk with an image file and Timing Save or Real Time Save might be a better option than Primo Cache. Alternatively, have VMs and databases on separate volumes and only cache the database one, if you're not doing so already.

[rit]

The aim is to improve the performance of the VMs so nether option is an option and this product does greatly improve a PC environment that is being used to run many different VMs. Using a RAMDISK does not work as the VM images I work with are about 2TB in size. And spliting the VMs between general and data is not a solution as that defeats the whole point of caching things

The issue comes when suspending a VM as this causes the memory image to be written back to disk so many GBs as fast as possible. As the cache software will accept this into its deferred block queue and so get overloaded.

[Support]

Well, in v2.0.0, you may try the defer-write mode buffer/average