SMR Tech HDD
SMR Tech HDD
2, 2015
• SHIMOMURA Kazuhito
The information infrastructure society has been undergoing further major changes in recent
years. Whereas most digital data were formerly created by individuals using PCs, the progress of
social networks has led to the generation of huge volumes of data on a daily basis from smart-
phones and other portable terminals. Furthermore, the data produced by various sensors in the
Internet of Things (IoT) has also been rapidly increasing. The total volume of data generated in the
world in 2020 is projected to reach 44 zettabytes (ZB; 1 ZB = 1021 bytes), 10 times the volume in
2013. Larger capacity hard disk drives (HDDs) are therefore required in the market to store these
huge volumes of data.
To address this demand, Toshiba has been developing shingled magnetic recording (SMR) tech-
nologies. SMR HDD record data in a narrower track pitch than conventional HDDs by overwriting
a part of the adjacent track that has been recorded, like shingling a roof, using a read/write head
of conventional size. We are working to realize new technologies for SMR that strengthen its
advantages and compensate for its weak points.
33
TOSHIBA REVIEW GLOBAL EDITION Vol. 1, No. 2, 2015
the write element of a head, W, is designed to be nar- overlapping part of the previously written track, N.
rower than the track pitch, TP (Figure 2). This means a narrower TP and thus allows for a higher
The recording capacity of a CMR HDD is a func- TPI without reducing the width of the head. Because
tion of a bits-per-inch (BPI) density in the down-track the tracks partially overlap similar to roof shingles,
(circumferential) direction and a tracks-per-inch (TPI) this approach is called shingled magnetic recording
density in the cross-track direction. In order to increase (Figure 4). Although both CMR and SMR have a
recording density, the width of a write element and the restriction for TP due to the width of the read element,
TP of a disk must be reduced (Figure 3). For this pur- SMR provides far greater design freedom than CMR
pose, the magnetic characteristics of both of them need does.
further improvement.
However, at present, the width of a write element is 3. Data recording
as narrow as approximately 50 nm, and it is becoming CMR temporarily buffers data from a host into mem-
difficult to further reduce its width without affecting ory. Regardless of whether writes are addressed to ran-
recording capability. HDD manufacturers have been dom or sequential locations on a disk, the controller
researching innovative technologies to break this limit, firmware reorders the writes in such a manner as to
but it requires much time and cost to develop key com- minimize the overall access time and then performs
ponents such as heads and disks. writes to the disk (Figure 5)
Here comes SMR, which writes a new track, N+1, In SMR, data are sequentially written to shingled
tracks just as the roof of a house is shingled. Therefore,
SMR does not have the random-access performance of
Track CMR. To resolve this issue, SMR HDDs need a special
N
Guard band writing procedure. Specifically, an SMR disk provides a
N+1 TP media cache (MC) and bands of shingled tracks. Bands
are defined as a unit consisting of shingled tracks that
N+2 W are sequentially written. There is a guard band between
contiguous bands.
Write element Read element
Upon a write request from a host, an SMR HDD tem-
Figure 2. Schematic diagram of data tracks, write element, porarily writes the buffered data to a disk area called an
and read element of conventional magnetic recording (CMR) MC. The addresses of data in the MC are recalculated
HDD. so that the data can be written to sequential locations
CMR has guard bands between adjacent tracks. The width of the
write element is narrower than the track width. in shingled bands. To rewrite previously written data,
an SMR HDD reads old data from a band and merges
it with new data in the MC and then writes it back to a
Track
new band (Figure 6).
N
Guard band In order to improve write performance, we have also
N+1
TP
developed an algorithm that directly writes data to
N+2
bands without involving the MC when a write request
N+3
for sequential data from a host is larger than a certain
N+4 size.
Write element Read element
N Track
N+1 Host
N+2 TP
N+3
N+4 Buffer memory
34
TOSHIBA REVIEW GLOBAL EDITION Vol. 1, No. 2, 2015
HDD
Fast
Write performance (random data)
Disk
Unwritten area Written data
Newly written data
Guard band
Band
Host
Slow
Buffer
memory MC
Written data New band
0 20 40 60 80 100 120
Time (min)
Figure 6. Data writing procedure in SMR HDD. Figure 8. Random data writing performance of SMR HDD.
In response to write requests from a host, the SMR HDD receives the The random-write performance of the SMR HDD varies, depending
write data with a buffer memory and copies it to the MC temporarily. on the amount of free space available in the MC.
Then, the write data is written in a shingle pattern to new bands.
SMR HDD
Fast
D0
N-2
CMR HDD D0
R Writing
is
TP N-1
O VR E stopped.
D0
Slow
E
W
Erasure width
Previously written track N
0 20 40 60 80 100 W
E Write element
Outer tracks Inner tracks
Track being written
Percentage of disk area written (%) E
TP = 2×D 0+O VR+(R+E)
Figure 7. Comparison of sequential-write performance of
CMR and SMR HDDs. Figure 9. Track pitch design with emphasis on reliability.
The SMR HDD has a sequential-write performance equivalent to that The track pitch is carefully determined to prevent accidental data
of the CMR HDD. erasure in the event of an external disturbance.
35
TOSHIBA REVIEW GLOBAL EDITION Vol. 1, No. 2, 2015
erasure that occurs at the edge of the write element. command standards such as Zone Block Command
D 0 must be reduced to decrease TP and thereby (ZBC) and Zone-device ATA Command (ZAC) stan-
achieve a high TPI. On the other hand, too small a D0 dards.
value would cause an SMR HDD to stop writing data There are three types of firmware for SMR: drive-
frequently, degrading write performance. Our SMR managed (DM), host-aware (HA), and host-managed
HDD incorporates innovative solutions for making (HM). Drive-managed SMR needs to run with an exist-
full use of the advantage of the SMR HDD in order to ing filesystem. While a host can handle a drive-managed
achieve even higher TPI with a minimal effect on write SMR HDD in the same manner as a CMR HDD, its
performance. performance varies depending on the host’s access con-
ditions and is hard to predict. In contrast, host-managed
6. Types of firmware and industry standardiza- SMR allows a host to manage the operation of HDDs
tion in a manner suitable for the characteristics of SMR and
Although, depending on data characteristics, SMR thus delivers stable performance. For hosts supporting
HDDs do not always exhibit performance comparable host-managed SMR, new technologies are being devel-
to that of CMR HDDs, SMR will certainly provide the oped to make it possible for filesystems and kernels to
capacity growth spurt. Studies on the utilization of run SMR HDDs in a desired manner. Host-aware SMR
SMR HDDs have begun not only at individual HDD is not purely driver- or host-managed, but has charac-
manufacturers but also across the entire industry sector teristics of each (Table 1).
involved in information infrastructure, including stor-
age system manufacturers, large-scale data centers, and 7. Conclusion
other users. More and more SMR HDDs will be available as high-
Furthermore, we have been engaged in the standard- er-capacity HDDs become necessary to meet customer
ization of SMR together with other companies. The and application needs.
International Committee for Information Technology CMR HDDs have been dominant for the past few
Standards (INCITS) has the T10 Technical Committee decades. As SMR HDDs gain wider uptake, changes will
responsible for the Serial Attached SCSI (*1) (SAS) stan- take place in the design of such key components as mag-
dards and the T13 Technical Committee responsible for netic heads and recording media. The host filesystem
the Serial Advanced Technology Attachment (SATA) will also be optimized to obtain the best performance
standards, each of which has proposed SMR-specific from SMR HDDs.
SMR HDDs will bring about a paradigm shift, involv-
ing system manufacturers and users, and find wide-
Table 1. SMR firmware categories
spread use in information infrastructure across which a
Category Characteristics huge volume of data flows.
We will incorporate the SMR technology into next-
The host can handle a CMR HDD in the same manner
Drive- as an SMR HDD. However, the HDD performance sig- generation high-capacity HDDs.
managed nificantly varies according to the host's access condi-
(DM) tions. SMR HDDs are downward compatible with CMR
HDDs. Reference
(1) Turner, V. et al. “The Digital Universe of Opportunities:
The host can optimize write operations using com-
Rich Data and the Increasing Value of the Internet of
mands specifically designed for SMR HDDs. Even in the
event of an SMR command is under violation, the SMR
Things.” IDC Website. Accessed May 25, 2015. http://idcdoc-
Host-aware HDD accepts the command and performs a possible serv.com/1678.
(HA) operation. The HDD performance sometimes varies
significantly according to the host's access conditions.
Host-aware SMR HDDs are downward compatible with
CMR HDDs.
The host uses SMR-specific commands to optimize
Host- write operations. The SMR HDD rejects an SMR com- SHIMOMURA Kazuhito
managed mand that violates to the specification. The write per-
(HM) formance can be estimated to some extent. SMR HDDs Chief Specialist. HDD Products Engineering
are not downward compatible with CMR HDDs. Department 1, Storage Products Design & Pro-
duction Division, Semiconductor & Storage
Products Company. He is engaged in the design
(*1) SCSI: Small Computer System Interface of HDD products.
Note:
This publication may include Toshiba’s original estimation, and/or data based on Toshiba’s evaluation.
36