A Brief History of RAID Technology
Redundant Array of Independent Disks (RAID) technology was pioneered in the late 1980s by researchers at the University of California, Berkeley in partnership with IBM. The goal was to use multiple hard drives together to improve performance, capacity and reliability beyond what a single drive could deliver at the time.
Several RAID levels were defined, each with its own unique blend of striping (spreading data across drives), mirroring (duplicating data on separate drives) and parity (error detecting code) to achieve different optimizations. For example, RAID 0 used striping to boost speed, while RAID 1 used mirroring for fault tolerance.
Over the years, RAID became a staple technology to store and protect critical data on servers and high-end workstations. Today it continues to be relevant with faster, larger and cheaper drives available. The most commonly used levels today are RAID 1 and RAID 5.
What is RAID 1?
RAID 1 is the simplest form of RAID that provides redundancy through disk mirroring. It involves an exact copy (or mirror) of data from one drive to another. A basic RAID 1 setup requires just two physical hard drives – one being the primary data drive while the second drive maintains a live copy of the same data.
All read and write operations by the server are performed on the primary data drive. The RAID controller duplicates every bit of data onto the secondary mirrored drive in real-time. This ensures the second drive has an exact replica of the primary drive at all times.
If the primary data drive fails, the secondary drive can continue providing an identical copy of the data without service interruption. This makes RAID 1 very fault tolerant and suitable for mission critical storage needs. Rebuilding the failed primary drive involves simple data copying from the functioning secondary drive.
A RAID 1 array can also include more than two drives, with a single primary data drive and multiple secondary mirror drives for additional redundancy.
Advantages:
- Excellent fault tolerance and easy rebuild of failed drives
- 100% redundancy of data
- Can sustain multiple drive failures if array includes more mirrors
- No parity calculation overhead → Faster writes
Disadvantages:
- High disk capacity overhead since mirrors require as much space as primary data
- Rebuilding large drive arrays can take substantial time
- Typically requires identical drive sizes and types
Use Cases: Critical data, databases, virtualization, file/print servers
What is RAID 5?
RAID 5 provides data redundancy through distributed parity instead of mirroring. It requires a minimum of 3 physical drives to implement.
Data is split up into blocks which are then written across all the drives but one. The remaining drive is dedicated to storing parity information which can be used to reconstruct data in case of a drive failure. This parity data is distributed (striped) across all drives rather than being on a single drive.
For example, in a 3-drive RAID 5 array, drive 1 may store data blocks A1 and B1, drive 2 may store A2, B2 and parity block P12, and drive 3 stores A3, B3 and parity block P13.
If any single drive fails, the missing data blocks can be rebuilt using the surviving data blocks along with parity blocks from the other drives. This provides fault tolerance and redundancy for RAID 5 arrays. Adding more drives can further improve redundancy.
Advantages:
- Good read performance with multiple drives handling load
- Low capacity overhead unlike mirroring
- Can sustain single drive failure without data loss
- Cost-effective redundancy
Disadvantages:
- Slower write performance due to parity calculation overhead
- Rebuilding failed drive in large arrays is more complex
- Poor performance with multiple drive failures
Use Cases: File servers, data warehousing, medium budget business storage
Comparing Core Capabilities
| Metric | RAID 1 | RAID 5 |
|---|---|---|
| Redundancy method | Disk mirroring | Distributed parity |
| Min drives needed | 2 | 3 |
| Drive size flexibility | Requires identical sizes | Allows different sizes |
| Capacity overhead | 100% (copies require equal capacity) | Varies, as low as 1 drive worth (for parity) |
| Max drive failure tolerance | (N-1) mirrors | 1 drive (N=number of drives in array) |
| Rebuild complexity after failure | Simple recopy from existing mirror | Complex rebuild by recalculating missing data using parity data and remaining drives |
| Read performance | Excellent (mirrors share load) | Excellent (multiple drives share load) |
| Write performance | Excellent (simple parallel writes) | Moderate (parity calculation overhead for each write) |
| Cost efficiency | Low (high capacity overhead) | High (low redundancy overhead) |
As the table illustrates, while both RAID 1 and RAID 5 provide data redundancy, they differ significantly in their core approach, capabilities and use cases.
Performance Impact
One of the foremost reasons to setup a RAID array is to improve performance for storage intensive applications. The inherent parallelism offered by spreading data across multiple drives provides a boost for heavy workloads.
Read Performance
For predominantly read operations, both RAID 1 and RAID 5 demonstrate excellent performance scaling almost linearly with number of drives added.
In a RAID 1 array, identical copies of data stored on the mirrors simply multiples the read performance since the workload is shared across drives. 4 x 500 MB/s SATA SSD mirrors effectively provide nearly 2 GB/s reads in RAID 1 configuration.
Similarly, distributing data over multiple drives in RAID 5 ensures all drives participate in handling read requests in parallel. By striping data, multiple drives can be leveraged efficiently to significantly enhance throughput.
Write Performance
Write operations reveal key differences in how RAID 1 and RAID 5 achieve redundancy.
Every single write on the primary RAID 1 data drive has to be instantly duplicated onto the mirror drive with minimal overhead. So RAID 1 write speeds can match up to the drives‘ physical sequential write performance levels. With today‘s SSDs reaching upto 3.5 GB/s speeds, RAID 1 delivers blazing fast writes.
However, RAID 5 exhibits slower write speeds due to the computation (known as XOR calculation) needed to generate and update parity on each drive after every write. This extra calculation latency and the rotational gaps from using traditional hard drives limits practical write speeds to 100-200 MB/s. Modern RAID cards and SSDs have improved latency but still trail RAID 1.
So applications with heavy write operations would see far better performance with a RAID 1 implementation.
Rebuild Process and Fault Tolerance
Ever increasing drive capacities pose greater risks and downtime when rebuilding failed drives, making redundancy and rebuild characteristics even more critical.
RAID 1 Rebuild
The rebuild process for RAID 1 is straight-forward by simply copying all data from an existing mirror drive onto a replaced drive. Data integrity verification using checksums is done to ensure complete accuracy of the rebuilt mirror.
Depending on the size of drives, this could take several hours to days on average for today‘s large 10-16 TB drives. But applications can function normally during this time using the existing mirror.
In fact, RAID 1 can tolerate failure of multiple mirror drives as long as one drive with complete data exists, thanks to its complete redundancy. This makes larger arrays highly resilient to multiple simultaneous drive failures – a scenario that spells doom for RAID 5 arrays as we discuss next.
RAID 5 Rebuild
When a drive fails in a RAID 5 array, rebuilding requires a complex and lengthy drive replacement and data rebuild procedure:
- The failed drive is physically replaced with a new blank drive in the array.
- The entire RAID array enters degraded mode but remains available for use.
- Data and parity blocks are read from remaining drives to recalculate contents of failed drive. Parity helps rebuild missing data.
- Data is written to replacement drive to restore redundancy.
This comprehensive rebuild process puts substantial stress on the surviving drives while degrading performance. In larger arrays, massive amounts of data across multiple terabyte drives makes RAID 5 rebuilds extremely taxing with high risk of collateral drive failures.
Studies by storage vendors and cloud providers have shown rebuild times taking days for large arrays – with significantly greater chance for data loss disasters. Manageability also becomes a headache with frequent rebuilds needing attention.
Compare this to the relatively quick and safe mirrored drive rebuild with RAID 1 arrays. The contrast is quite stark for business critical systems needing stability.
Cost Considerations
While RAID 1 and RAID 5 redundancy have tangible differences, a key factor for most buyers is determining the cost effectiveness for desired storage capacity and performance. We analyze this across both upfront and operating costs.
Upfront hardware costs
As we‘ve discussed earlier, RAID 1 requires a minimum of 2 drives versus 3 drives for entry-level RAID 5. So RAID 1 has a lower starting hardware cost in scenarios where capacity requirements are lower.
However if storage needs grow substantially, RAID 5 becomes significantly more cost efficient. Due to its distributed parity approach, RAID 5 arrays can scale to higher capacities without doubling/tripling every drive like mirroring mandates. Adding drives to a RAID 5 simply adds one unit of capacity cost each time. RAID 1 adds 2X capacity cost per drive.
This causes RAID 5 to trend cheaper as arrays grow large. Though performance begins degrading beyond 8-10 drives for both RAID levels.
Operating costs
During the service life of a RAID array, operating costs add up in terms of factors like power usage, cooling needs and spare drives inventory.
RAID 1 again starts favorably thanks to lower drive counts. But frequent rebuilds of ever growing failed drives can quickly escalate costs of maintenance labor, spare parts stock and service disruptions from RAID 5.
Studies peg these operational costs to be 3-4X higher long term for large RAID 5 farms. The savings from lower initial hardware is often outweighed by backend expenses and risks.
So prudent businesses prefer RAID 1‘s operating model for critical systems despite marginally higher starting costs. The total cost of ownership is lower in most real world use cases.
Typical Use Case Examples
The strengths and weaknesses covered so far clearly underline the best suited applications for each RAID type:
RAID 1 Use Cases
- Database servers – need fast mirrored writes and rebuilds
- File and print servers – highly reliable and redundant storage is vital
- Mail servers – ensure zero email data loss
- Virtualization hosts – prevent VM downtime due to storage failures
- Critical application data – financial systems, healthcare records, CRM
RAID 5 Use Cases
- Archive storage – rarely written, but read often
- Media servers – video editing, resized on reads
- Code repositories – high uptime, heavy reads
- General purpose file sharing – affordable capacity
As a rule of thumb, RAID 1 fits best for mission critical systems that prioritize availability and performance. RAID 5 is preferred where capacity is paramount and redundancy needs permit slower rebuilds.
Conclusion
We‘ve explored the core technical differences between the two most popular RAID levels – RAID 1 and RAID 5. Both provide much needed data redundancy against drive failures which are inevitabilities with current hard drive technology.
While the surface goal of reliable storage is common, RAID 1 and 5 take divergent approaches with their mirroring versus distributed parity models. This manifests in very dissimilar feature sets when it comes to performance, fault tolerance, rebuilds, scalability and costs.
By understanding where each solution excels based on your workload patterns and budget, the choice becomes clear. OLTP databases shine with RAID 1‘s uncompromising speed and uptime capabilities. Hadoop clusters lean towards RAID 5 for oceans of slow changing data.
I hope this comprehensive RAID 1 versus RAID 5 comparison has demystified the two technologies to help decide the best option for your environment. Do share your thoughts or queries in the comments section!
