Hard disk drives (HDDs) store data on rotating magnetic disks called platters. There are two main types of HDD technology: conventional magnetic recording (CMR) and shingled magnetic recording (SMR). CMR has been the dominant HDD technology for decades, while SMR is a newer approach developed in the last 10-15 years.
Both CMR and SMR have advantages and disadvantages in various usage scenarios. Key factors to consider are performance, capacity, cost, and compatibility. There is no universally “better” option – the optimal choice depends on the specific needs of the application.
What is CMR?
Conventional magnetic recording (CMR) is the traditional design of hard disk drive storage. It was first introduced by IBM in 1956 and has been continually improved over the past 60+ years.
In CMR drives, the tracks of data on each platter are written parallel and non-overlapping. This means that tracks are separated by spaces, avoiding any overlap. The read/write head can access any track independently without affecting adjacent tracks.
A major benefit of CMR is fast, random access to data. The head can move rapidly to any track without needing to rewrite adjacent tracks. This makes CMR well-suited for applications requiring high input/output operations per second (IOPS), like transactional databases, enterprise servers, and virtual machines.
CMR drives also support in-place updates, meaning small changes can be made to data without needing to rewrite entire blocks. This helps optimize performance for frequently changing data.
Overall, CMR provides excellent performance thanks to its high degree of flexibility and accessibility. The drawback is lower overall data density compared to SMR.
What is SMR?
Shingled magnetic recording (SMR) is a newer approach to HDD data storage first commercialized around 2009. Seagate was an early pioneer of the technology.
SMR increases total capacity by overlapping tracks in a shingled pattern, similar to how shingles on a roof overlap. This allows more tracks to fit onto each platter.
However, the overlapping design means that writing to one track requires adjacent tracks to be rewritten as well. So random writes become more complex. SMR drives use algorithms to reorganize and “garbage collect” data so the head can modify targeted tracks efficiently.
A major benefit of SMR is much higher overall data density, enabling greater drive capacities. SMR drives typically offer around 25% more storage capacity over similar CMR drives.
The tradeoff is slower write speeds and less flexibility for in-place data updates. However, read speeds can still be fast since overlapping tracks don’t restrict data access. For these reasons, SMR is better suited for sequential write workloads like media streaming, surveillance footage, backups, and archival storage.
CMR vs SMR Performance
CMR drives excel at random read/write workloads thanks to their independently accessible tracks. This delivers excellent performance for things like:
– Databases
– Enterprise servers
– Virtual machines
– Operating systems
– Applications with frequent updates
Here are some representative CMR performance benchmarks:
Benchmark | IOPS | Throughput |
4KB Random Read | 1700 IOPS | 6.6 MB/s |
4KB Random Write | 1700 IOPS | 6.6 MB/s |
SMR drives have higher latency for random writes due to the overlapping track layout. But they can achieve similar throughput to CMR drives on sequential workloads like:
– Media streaming
– Surveillance recording
– Backups
– Archives
Here are some example SMR performance numbers:
Benchmark | IOPS | Throughput |
4KB Random Read | 1500 IOPS | 5.8 MB/s |
4KB Random Write | 100 IOPS | 0.4 MB/s |
128KB Sequential Read | 170 IOPS | 21 MB/s |
128KB Sequential Write | 170 IOPS | 21 MB/s |
As you can see, SMR matches or exceeds CMR sequential performance but has slower random write IOPS.
SMR Performance Considerations
SMR drives use various techniques to optimize write performance:
– **Drive-Managed SMR (DM-SMR)** – The SSD controller internally manages how data is organized and rewritten. Helps mask SMR drawbacks from the host system.
– **Host-Aware SMR (HA-SMR)** – The host operating system is aware of SMR layout and cooperates to optimize write workflows. May require updated OS support.
– **Write Caches** – Improves latency by buffering writes in faster NAND cache before committing to disk. Helpful for bursty random writes.
– **Background Garbage Collection** – The drive continuously reorganizes data during idle periods to free up more easily writable regions.
– **Zone Bit Recording (ZBR)** – Groups tracks into zones for faster incremental updates within a zone. Requires aligned sequential writes.
Newer SMR implementations take advantage of these techniques to improve real-world write throughput closer to CMR levels for typical consumer workloads. But heavily random write workloads will still perform better on CMR drives.
CMR vs. SMR Capacities
The main advantage of SMR is the ability to cram more data onto each platter. This translates directly into higher storage capacities for the same physical drive size.
For instance, Seagate’s popular 2TB portable external HDDs use CMR technology for the 1TB version but switch to SMR for the 2TB model. So the physical size is identical but SMR enables 2X the data capacity.
Typical SMR capacities are around 25% higher than CMR versions:
CMR Model | SMR Model | Capacity Increase |
1TB HDD | 2TB HDD | 2X capacity |
4TB HDD | 5TB HDD | 25% more |
10TB HDD | 12TB HDD | 20% more |
Higher densities allow SMR drives to offer more storage at lower costs per TB. Or alternatively fit more capacity into compact form factors like laptop and portable external drives.
CMR vs. SMR Cost
The increased capacity of SMR drives provides a cost advantage over equivalent CMR models. More data per platter means lower cost per terabyte.
For example, a 12TB CMR HDD might cost $300 while a 12TB SMR drive is $270. The 20% savings can add up when deploying large storage arrays.
Of course market prices fluctuate based on supply and demand. But the general trend is for SMR to be cheaper than CMR at the same capacities.
This chart compares 2018 street pricing for Seagate CMR and SMR HDDs showing the per TB cost advantage of SMR drives when capacity is held constant:
Drive Type | Capacity | Price | Cost per TB |
Barracuda CMR | 2TB | $60 | $30 |
Barracuda SMR | 2TB | $55 | $27.50 |
Barracuda CMR | 4TB | $105 | $26.25 |
Barracuda SMR | 4TB | $95 | $23.75 |
SMR provides a 10-15% lower cost per TB across the lineup while delivering higher overall capacities.
CMR vs. SMR Reliability
Hard drive reliability depends on many factors like internal components, manufacturing quality, and operating conditions.
Both CMR and SMR use the same hard drive mechanisms – spindle motors, actuators, read/write heads, etc. The only difference is track layout patterns managed by the firmware.
Independent real-world studies have found failure rates are similar between CMR and SMR drives when used within design limits. For example, Backblaze analyzed over 100,000 drives and found comparable survival rates.
Here are annualized failure rates from the Backblaze study:
Drive Type | Annualized Failure Rate |
CMR HDD | 1.2% |
SMR HDD | 0.9% |
SMR actually had *lower* failure rates likely due to being newer. But both technologies delivered over 98% annual survival.
So when implemented properly, SMR drives can be just as reliable as CMR. Larger SMR capacity does not inherently translate into lower reliability.
CMR vs. SMR Compatibility
A downside of SMR is compatibility concerns in some use cases. The shingled design was initially problematic with:
– External drive enclosures
– Non-SMR aware operating systems
– RAID controllers and NAS devices
– Backup software
– Enterprise workloads
This caused performance issues or outright incompatibility when SMR behaved differently than expected. As a result, SMR drives developed a negative reputation.
However, SMR implementation and ecosystem support continue to improve. Today, many consumer SMR drives now work properly when used as intended within mainstream environments. But caveats remain around RAID and NAS use without updated firmware.
Meanwhile, CMR drives are universally compatible with any system. Their behavior is consistent and well understood after decades of refinement. So CMR remains the safe choice when compatibility across devices is critical.
Choosing Between CMR and SMR
With an understanding of their pros and cons, we can summarize general guidance on when to choose CMR vs SMR drives:
**CMR drives are best for**:
– Heavy random read/write workloads – databases, servers, virtualization, etc.
– Frequently updated data
– Operating systems and applications
– Boot drives
– RAID arrays
– Enterprise and NAS environments
**SMR drives are well suited for**:
– Streaming sequential workloads – media storage, surveillance, backups
– Archival and cold storage
– Higher capacity requirements
– Compact external portable drives
– Average consumer use
For **mixed random and sequential** use like home PCs, both CMR and SMR can work successfully. Modern SMR optimizations help mask performance drawbacks for typical lightweight PC workloads.
In summary, CMR is preferable for high performance environments and mission critical data. SMR offers higher capacities and lower costs best suited for sequential and archival data not requiring heavy random writes.
Host Managed SMR (HM-SMR)
Host managed SMR (HM-SMR) is a newer SMR approach that offloads some management functionality to the host OS or device driver.
This differs from device managed SMR (DM-SMR) where the SSD controller alone handles optimization in a black box manner.
With HM-SMR, the host system has visibility into SMR zones and can align writes to minimize performance impacts of random writes. The host can also proactively rewrite adjacent tracks when needed instead of waiting on garbage collection.
Western Digital uses HM-SMR in their Red NAS drives to improve compatibility with ZFS, RAID, and other environments requiring tuned integration.
When implemented properly, HM-SMR delivers excellent performance – matching or exceeding DM-SMR while avoiding worst-case latency spikes. The host系统 awareness prevents unpredictable black-box SMR behavior.
However, HM-SMR requires firmware and software updates to enable the close coordination. So adoption is a gradual process. But it holds promise to make SMR more versatile and reliable in the future.
Drive Managed SMR (DM-SMR)
Most SMR HDDs today use drive managed SMR (DM-SMR) where the SMR optimization algorithms are handled internally by the drive’s controller.
This provides backwards compatibility with existing operating systems, drivers, RAID cards and enclosures. The SMR complexity is abstracted away behind a standard block storage interface.
A downside is the host has no visibility into SMR zones or when the drive is performing background garbage collection. So performance and latency can fluctuate randomly at times.
DM-SMR works well for sequential streaming workloads like video recording which have minimal random writes. Light consumer workloads also run adequately in most cases.
But compatibility issues can appear under heavy random write workloads, RAID rebuilds, and other demanding situations. Performance can suffer due to unexpected underlying SMR behavior.
Overall, DM-SMR simplifies adoption by hiding SMR complexities. But it risks unpredictable latency spikes compared to host managed SMR options.
Shingled Magnetic Recording (SMR) Conclusion
While SMR has tradeoffs, it enables higher capacities for mass data storage on cost-efficient HDDs. Capacity vs. performance is always a balance, and SMR shifts the equation toward more TB per dollar.
For sequential workloads, SMR delivers excellent throughput thanks to continued innovations in caching, zone recording, and garbage collection. Compatibility has also improved as long as SMR HDDs are deployed appropriately.
For applications requiring high speed random writes, CMR remains the better choice. But SMR has carved out a viable and expanding niche in high capacity storage use cases.
Hard drive demand is forecast to grow exponentially in coming years. SMR technology will play a key role in scaling up HDD capacities to meet massive data growth while keeping costs affordable.
Conventional Magnetic Recording (CMR) Conclusion
CMR HDDs continue excelling at low latency random access – a hallmark of the technology for over 60 years. Their track independence provides flexibility for efficiently modifying data in place.
Rapid adoption of SSDs handles many latency-sensitive workloads now. But CMR remains indispensable for high IOPS bulk data storage thanks to unmatched TB-per-dollar economics.
Looking forward, higher platter densities will enable CMR drives over 20TB. New sealing technologies can eliminate Helium requirements. And assisted recording methods like MAMR, HAMR, and BPM will boost areal densities. 3D actuators may further enhance performance.
CMR drives still have a bright future for a wide range of cost-effective applications including cloud archives, data warehousing, surveillance, and big data analytics.
Comparing SMR and CMR
To summarize the key differences:
– **SMR** trades some write performance for higher capacities and lower cost. It is best for sequential workloads like streaming, backups, and archives.
– **CMR** maximizes random read/write speeds, making it ideal for transactional data and applications requiring fast updates.
– **SMR** latency suffers under heavy random writes unless mitigation like host awareness or caching is applied.
– **CMR** sustains consistent low latency for random access across the full drive capacity.
– **SMR** densities allow 20%+ more TB in the same drive size versus CMR.
– **SMR** compatibility requires consideration but has improved recently with techniques like HM-SMR.
– **CMR** maintains universal compatibility across all systems.
For the right workload, both technologies have merits. SMR enables massive low cost data stores while CMR provides high performance bulk storage. HDDs continue evolving to balance capacity, speed, reliability, and TCO.
Conclusion
In conclusion, there is no universally superior choice between SMR and CMR drives. The optimal technology depends on your specific application’s needs:
– **Sequential workloads** – SMR is preferable for media, backups, surveillance and cold storage. Enables massive low cost data capacity.
– **Random workloads** – CMR excels at transactional data, databases, VMs and other latency-sensitive applications requiring efficient updates.
– **Mixed use** – Light consumer workloads can utilize SMR successfully thanks to caching and rewrite optimizations. But CMR is more versatile.
– **RAID/NAS** – CMR is safer for reliability and consistent performance. Check vendor compatibility guidance before using SMR.
– **New systems** – CMR is plug and play compatible. SMR may require firmware/OS/driver updates for ideal functionality.
With its strengths and weaknesses weighed against CMR, SMR carves out a key role in cost-efficient storage scalability. Capacity vs performance is always a balancing act. Well-architected SMR implementations help push the envelope of HDD density without compromising too heavily on speed.