Location aware cache algorithm
Publication Date: 2018-Jan-16
The IP.com Prior Art Database
Location aware cache algorithm Previous implementation about SSD(solid state disk) cache shows as Figure 1:
• Downgrade SSD cache driver “Cache IO(input/output) distribution layer” to HBA(Host Bus Adapter).
• Create a specific Write-Hit command to write HDD(HardDiskDrive) and SSD in the “Write Hit” case.
• Create a specific Read-Miss command to read HDD and write SSD in the “Read Miss” case.
• Each command can read/write HDD data and update SSD cache within one IO operation. SSD cache content is always sync with HDD data. There is no need to choose to write through or write back.
The problems with previous implementation shows as Figure 2:
• It randomly chooses target SSD on different HBAs
• It can’t fully utilize the offload technology
Figure 2 Our solutions: To keep the advantage of "update SSD/HDD within one IO", need to make HDD data and SSD cache within same HBA as much as possible. To do this, we'd modify host Cache algorithm layer:
• Add a local bit in the cache calculation data structure
• Decide “hot” data to SSD cache
• Decide in local or remote SSD cache: ‘Write-frequency’ hot data to local SSD cache (make use of ‘write-hit’ in one
IO) Otherwise to any available SSD cache
• Define a new location aware cache algorithm to improve the cache efficiency for bandwidth and speed
• For read-more and write-more hot data, it should be put into the local HBA and fully utilize the offload commands
• For write-only and read-less hot data, it should be put into the local HBA and fully ut...