SSD Types and Their Impact on rXg Performance (2026, February)

This document provides a comprehensive analysis of solid-state drive technologies and their suitability for rXg network gateway deployments. The rXg platform performs database operations, extensive logging, session management, and real-time traffic analysis—workloads that benefit significantly from appropriate storage selection. The choice between consumer, prosumer, and enterprise SSDs can dramatically affect system reliability, performance consistency, and total cost of ownership.

---

For rXg deployments, NVMe Gen4 represents the optimal balance between performance, cost, and thermal characteristics. Gen5 drives offer diminishing returns for gateway workloads while introducing higher power consumption and heat generation. SATA SSDs should only be considered for legacy hardware constraints or extremely budget-conscious deployments where performance is secondary.

For rXg workloads: TLC NAND provides the best balance. QLC should be avoided for primary system drives due to the write-intensive nature of logging and database operations. Enterprise drives often use high-quality TLC or eMLC (enterprise MLC) with additional overprovisioning.

Modern SSD controllers come from several manufacturers with distinct characteristics:

The Phison E25 (4-channel) and E18 (8-channel) dominate the prosumer market. The E21T provides excellent budget performance with HMB support. Newer Phison E31T and upcoming E37T target Gen5 with improved power efficiency under 2.3W active power consumption.

The SM2262EN and SM2264 power many reliable drives. The SM2504XT represents the latest Gen5 DRAM-less controller on TSMC's 6nm process, consuming only 2.4W active while delivering up to 11,500 MB/s throughput.

Samsung's proprietary Pascal and Elpis controllers are vertically integrated with their V-NAND, offering tight optimization but at premium pricing.

---

SSD controllers use DRAM to cache the Flash Translation Layer (FTL) mapping table, which tracks the logical-to-physical (L2P) address translation. DRAM operates approximately 100 times faster than NAND flash for random access operations. A typical 1TB drive requires approximately 1GB of mapping table data (1MB per 1GB of capacity).

1. Faster random I/O: Direct L2P lookups without NAND access

2. Consistent latency: No FTL lookup delays during mixed workloads

3. Lower write amplification: More efficient garbage collection coordination

4. Better sustained performance: Maintains speed during extended operations

DRAM-less SSDs eliminate onboard DRAM to reduce costs and power consumption. Modern DRAM-less drives use Host Memory Buffer (HMB), a feature introduced in NVMe 1.2 that allows the SSD to borrow a small portion of system RAM (typically 64MB) for FTL caching.

According to Western Digital research, 95% of all typical activities experience no performance loss with HMB. With 64MB of HMB allocation, a 100% cache hit rate can be achieved for common workloads. However, HMB introduces additional latency due to PCIe bus traversal and host memory access coordination.

DRAM-less drives exhibit performance degradation under:

1. Sustained sequential writes exceeding SLC cache: Users report speeds dropping to 1 MB/s during 30GB+ continuous writes as the drive falls back to direct TLC/QLC writes

2. Heavy mixed random I/O: Random read/write patterns with high queue depths

3. Drive capacity above 80% full: Reduced space for SLC cache and garbage collection

4. Systems with limited RAM: HMB competes with system memory requirements

---

3. rXg Workload Characteristics and SSD Requirements

The rXg platform generates specific storage workload patterns that inform SSD selection:

Database Operations:

PostgreSQL database for session management, user authentication, captive portal data, and configuration storage. This creates mixed random read/write patterns with periodic VACUUM operations that trigger substantial write activity.

Continuous syslog writes, RADIUS accounting, DHCP lease tracking, and network flow data. These workloads are predominantly sequential writes with periodic random reads for queries and reporting.

Operating system operations, temporary files, and application data. Generally light but continuous background activity.

Write Amplification Factor (WAF) significantly impacts SSD longevity in rXg deployments. WAF represents the ratio of actual NAND writes to host writes—a WAF of 3 means 1GB of host data results in 3GB of NAND writes.

1. Small random writes: Database commits and log entries

2. Drive capacity utilization: Higher fill levels increase garbage collection overhead

3. NAND type: QLC has higher inherent WAF than TLC

4. Lack of TRIM support: Filesystem and hypervisor compatibility issues

1. Maintain drive capacity below 75% for adequate overprovisioning headroom

2. Enable TRIM/DISCARD at filesystem and database levels

3. Select drives with higher native overprovisioning (enterprise drives typically use 7-28%)

4. Consider separate drives for logs (high write) and database (mixed I/O)

For a typical rXg deployment:

Minimum TBW Recommendation: For a 5-year deployment with moderate logging, target at least 600 TBW for a 1TB drive (accounting for 3x WAF). High-traffic deployments with extensive flow logging should target 1,200+ TBW.

---

Network gateways operate 24/7 and may experience unexpected power events. Power Loss Protection (PLP) prevents data corruption during sudden power loss by using onboard capacitors to complete in-flight writes to NAND.

Consumer SSDs (including prosumer models) typically offer firmware-based power loss immunity that protects data at rest from corruption during new write operations. This is NOT the same as enterprise PLP, which uses hardware capacitors to:

1. Complete all pending writes from volatile caches

2. Flush the mapping table to NAND

3. Ensure metadata consistency for recovery

Without proper PLP, a power loss during database commits or log writes can result in filesystem corruption, PostgreSQL recovery failures, or complete drive failure. For rXg deployments without UPS protection, enterprise PLP is strongly recommended.

4.2 PLP Implementation Levels

Enterprise drives use tantalum polymer or aluminum capacitors rated for 10-50ms of backup power—sufficient to flush all buffers and finalize the FTL state.

---

Use Case: Small deployments, home lab testing, budget-constrained environments with UPS protection.

Offers the best price-to-performance ratio with Phison E21T controller and Micron 176-layer TLC. The 600 TBW endurance is adequate for light rXg deployments. Avoid the QLC variant.

1. No hardware power loss protection—requires UPS

2. Performance degradation under sustained writes (log spikes)

3. Lower endurance may require replacement within 3-4 years under heavy use

4. Limited or no enterprise support/warranty handling

Use Case: Production rXg deployments, SMB environments, situations requiring reliable performance without enterprise pricing.

The Crucial T500 delivers exceptional value with Micron's 232-layer TLC NAND and Phison E25 controller. Key advantages:

1. Superior efficiency: 517 MB/s per watt vs competitors' ~450 MB/s/W

2. Larger SLC cache: ~720GB dynamic cache vs Samsung's 226GB

3. Higher MTBF: 2.0 million hours (industry-leading for consumer class)

4. Competitive pricing: Generally less expensive than Samsung 990 Pro at equivalent capacities

The 2TB capacity provides headroom for logging growth and reduces drive utilization percentage, improving sustained performance and longevity.

Choose Samsung when brand support relationships exist or when the absolute fastest random read performance at high queue depths is required. The 990 Pro edges ahead at QD32 random reads (1,350K vs 1,180K IOPS) but typically commands a premium.

Use Case: Large-scale deployments, high-availability requirements, environments with strict compliance requirements, deployments without UPS protection.

The Micron 7450 Pro offers the optimal balance for rXg enterprise deployments:

1. Full hardware PLP: Tantalum capacitors protect all in-flight data

2. 1 DWPD endurance: 3,504 TBW over 5 years, ample for intensive logging

3. Consistent latency: Enterprise-grade controller with deterministic performance

4. TCG Opal 2.0: Hardware encryption for compliance requirements

5. Strong write performance: 5,300 MB/s sequential write outpaces competitors

For deployments with extremely heavy write loads (extensive flow logging, high-traffic environments), the D7-P5620's 3 DWPD rating provides 8,760 TBW over 5 years—nearly 2.5x the 7450 Pro's endurance.

For deployments requiring extended log retention or large database storage, the D5-P5336 offers up to 122.88TB in a single drive. The 0.58 DWPD is optimized for read-intensive workloads with moderate writes—suitable for log aggregation and analytics.

Intel discontinued Optane production in 2022, but remaining inventory and secondary market units offer exceptional characteristics for write-intensive database workloads:

Optane provides approximately 10x lower latency than NAND-based SSDs and maintains consistent performance under mixed random workloads. For rXg deployments with extremely demanding PostgreSQL workloads, a small Optane drive dedicated to database files can dramatically improve transaction throughput.

Caution: Optane is end-of-life with no future development. Use only where existing inventory is available and plan for eventual migration to NAND alternatives.

---

6. Configuration Recommendations

For most rXg deployments, a single high-quality prosumer or enterprise SSD is sufficient:

For optimal performance and reliability in larger deployments:

Software RAID 1 (mdraid or ZFS mirror) provides redundancy against drive failure. Use enterprise drives with full PLP to prevent array corruption during power events.

/dev/nvme0n1p1 /var ext4 defaults,noatime,discard 0 2

1. `noatime`: Reduces unnecessary write operations for access time updates

2. `discard`: Enables continuous TRIM for optimal garbage collection (alternatively, use periodic `fstrim` via cron)

1. Enable `wal_sync_method = fdatasync` for optimal write performance

2. Consider separate tablespace on high-endurance drive for heavy-write tables

3. Set `checkpoint_completion_target = 0.9` to spread checkpoint writes

---

Monitor these critical SMART attributes for SSD health:

bash
smartctl -a /dev/nvme0n1
nvme smart-log /dev/nvme0n1

Replace drives proactively rather than waiting for failure. The "Percentage Used" SMART attribute provides a direct indicator of remaining endurance.

---

Home Lab / Testing: Silicon Power UD90 1TB with UPS protection

Small Business (1-100 users): Crucial T500 1TB with UPS protection

Medium Business (100-1000 users): Crucial T500 2TB with UPS protection

Large Enterprise (1000+ users): Micron 7450 Pro 1.92TB with or without UPS

High-Availability / Compliance: Dual Micron 7450 Pro in RAID 1 with UPS and generator backup

---

- [TechTarget: Compare DRAM vs. DRAM-less SSDs](https://www.techtarget.com/searchstorage/feature/Compare-DRAM-vs-DRAM-less-SSDs-for-cost-performance)

- [Phison Blog: DRAM or Not?](https://phisonblog.com/dram-or-not-the-difference-between-dram-and-dram-less-ssds-and-why-it-matters/)

- [Phison Blog: Host Memory Buffer](https://phisonblog.com/host-memory-buffer-2/)

- [Tom's Hardware: Best SSDs 2026](https://www.tomshardware.com/reviews/best-ssds,3891.html)

- [PCWorld: Host Memory Buffer Technology](https://www.pcworld.com/article/784380/host-memory-buffer-hmb-the-dram-less-nvme-technology-thats-making-ssds-cheaper.html)

- [Serve The Home: HMB NVMe SSDs](https://www.servethehome.com/what-are-host-memory-buffer-or-hmb-nvme-ssds/)

- [PLOS ONE: HMB in DRAM-less NVMe SSDs](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0229645)

- [Kingston: SSD Power Loss Protection](https://www.kingston.com/en/blog/servers-and-data-centers/ssd-power-loss-protection)

- [ATP: SSD TBW and DWPD Endurance](https://www.atpinc.com/blog/ssd-tbw-dwpd-endurance)

- [Virtium: DRAM vs DRAM-less Selection](https://www.virtium.com/technology-updates/how-to-select-between-dram-vs-dram-less-ssds/)

- [StoredBits: Samsung 990 Pro vs Crucial T500](https://storedbits.com/samsung-990-pro-vs-crucial-t500/)

- [TweakTown: Silicon Motion SM2504XT Review](https://www.tweaktown.com/reviews/11137/silicon-motion-sm2504xt-controller-best-dramless-ssd-ever/index.html)

- [StoredBits: Silicon Power UD90 Review](https://storedbits.com/silicon-power-ud90-review/)

- [ScyllaDB: Intel Optane Analysis](https://www.scylladb.com/2017/09/27/intel-optane-scylla-providing-speed-memory-database-persistency/)

- [Intel: Optane SSD Database Performance](https://cdrdv2-public.intel.com/762400/Intel_Optane_db_perf_Jan-2023.pdf)

- [Wikipedia: Write Amplification](https://en.wikipedia.org/wiki/Write_amplification)