Fig 1.
Architecture of DRAM-less SSD supporting HMB.
Table 1.
Tested SSDs.
Table 2.
Host PC environment.
Fig 2.
I/O performance comparison of DRAM-less SSDs (SSD-A, SSD-B, SSD-C) and SSDs with internal DRAM (SSD-D, SSD-E, SSD-F).
(a) LIGHT, sequential read, (b) LIGHT, sequential write, (c) LIGHT, random read, (d) LIGHT, random write, (e) HEAVY, sequential read, (f) HEAVY, sequential write, (g) HEAVY, random read, and (h) HEAVY, random write.
Table 3.
Two workloads by fio.
Fig 3.
Some scenarios by read requests.
Fig 4.
An algorithm for testing the existence of a read cache.
Fig 5.
Test results for the existence of a read cache in the HMB of various SSDs.
(a) SSD-A, with HMB, (b) SSD-B, with HMB, (c) SSD-C, with HMB, (d) SSD-A, without HMB, (e) SSD-B, without HMB, and (f) SSD-C, without HMB.
Fig 6.
Some scenarios by write requests.
Fig 7.
An algorithm for testing the existence of a write buffer.
Fig 8.
Test for existence of write buffer in the host memory buffer.
(a) SSD-A, with HMB, (b) SSD-B, with HMB, (c) SSD-C, with HMB, (d) SSD-A, without HMB, (e) SSD-B, without HMB, and (f) SSD-C, without HMB.
Fig 9.
An algorithm for testing the existence of a mapping table cache.
Fig 10.
Test for existence of mapping table cache in HMB.
(a) SSD-A, (b) SSD-B, and (c) SSD-C.