Hystor
Hystor is a research block-storage system that combines separate SSD and HDD devices using cost-aware placement and optional write buffering.
Its frequency/request-size proxy prioritizes costly blocks; filesystem metadata receives separate treatment. After promotion, SSD timings no longer reveal the block's HDD latency. [1]
The remap area retains selected blocks; a separate write-back area buffers incoming writes. HDD home locations remain reserved, but dirty SSD data cannot simply be discarded. [1]
These policies do not establish crash durability. For comparison, Linux dm-cache separately specifies cache modes, mapping metadata and FLUSH/FUA commits; it is not a Hystor implementation. [2]
The linked constructed example isolates the observation problem with fixed, invented costs. It does not reproduce Hystor's policy or measurements. [3]
Relations
- related
- SSHD (hybrid drive), Solid-state drive (SSD), Hard disk drive (HDD), Write-back cache, Cache admission, SSD mapping cache
Limitations
- The evaluation used Linux 2.6.25.8 and Ext3 with historical SSD/HDD hardware; it does not establish present-day performance. [1]
Citations
- Chen, Koufaty and Zhang. Hystor. ICS 2011. DOI 10.1145/1995896.1995902; sections 3, 5 and 7.
- Linux kernel documentation: dm-cache, operating modes and on-disk metadata.
- Hesela: moving a block changes the latency you observe
Sources & caveats
- Primary-source synthesis and a separately labeled synthetic example; no hardware benchmark or power-loss test reproduced.
- Evaluation devices: Intel X25-E 32 GB SSD and Seagate Cheetah 15k.5 SAS HDD; section 7.1.