Page 1 of 1
Questions/Feature request
Posted: Mon Jul 11, 2016 4:59 pm
by npelov
Hi,
I notice there is a lot of data written to L2 cache. Since L2 is SSD I would like to minimize writing.
My questions are:
1. When PrimoCache decides that the data is good enough to be written in L2 cache
2. Does L2 cache contain frequency information?
Feature requests are based on answers of above, but it would be nice to have a limit of how frequently must a sector must be accessed to be worth writing to L2 cache. This way one time reads won't go to L2 - less writing.
Re: Questions/Feature request
Posted: Tue Jul 12, 2016 1:59 am
by idefix44
For some options, avanced users can use the CLI...
Re: Questions/Feature request
Posted: Tue Jul 12, 2016 2:42 pm
by Support
PrimoCache was able to choose data to L2 cache by frequency. However, it seems that the result was not good and lots of customers complained that the speed of populating L2 data was too slow. So now PrimoCache removes this feature and don't filter read-data to L2 cache. As you may know, it occupies the memory resource and cpu load to maintain the frequency information.
As you mentioned, some data is just one time read and it's a waste to put them into L2. We're developing a mechanism to filter these useless data.
Thanks.
Re: Questions/Feature request
Posted: Wed Jul 13, 2016 6:37 pm
by npelov
support wrote:PrimoCache was able to choose data to L2 cache by frequency. However, it seems that the result was not good and lots of customers complained that the speed of populating L2 data was too slow.
That's why I suggested making it option in cache configuration - so everyone can choose their preferences. For me it's most important to minimize L2 writing. With spinning disks it's not the amount of data that's important - it's minimizing the head travel. So caching the file system (frequently accessed) and most used sectors will allow the disk to read only file data. Having the disk defragmented now and then will make the file data sequential read which is 100s time quite faster than random reads.
support wrote:As you mentioned, some data is just one time read and it's a waste to put them into L2. We're developing a mechanism to filter these useless data.
Well, I'm just saying my concerns. I'm sure you will choose the best way to implement this.