image_pdfimage_print

TL;DR: For storage networks, use 9,000 bytes (jumbo frames) for NVMe/TCP and iSCSI and standard 1,500 bytes for management traffic.

In the world of networking, the term Maximum Transmission Unit (MTU) is critical when optimizing storage networks for performance and reliability. MTU size refers to the largest data payload that can be transmitted in a single packet without fragmentation. For storage professionals—especially those deploying NVMe/TCP or managing high-speed networks—understanding MTU sizing, jumbo frames, and NVMe/TCP MTU requirements is essential.

What Is MTU Size?

MTU represents the maximum payload size of a network packet that can be transmitted across a network. It specifies the largest data size that can be encapsulated within a packet and sent from one device to another without being divided into smaller fragments. Maximum transmission unit size is measured in bytes, and it includes both the data payload and the packet headers.

Larger MTU values (e.g., 9,000 bytes) are commonly called “jumbo frames” and allow more data per packet, which can reduce CPU overhead and improve throughput.

Maximum Transmission Unit Explained

Data sent across a network is broken into packets. Each packet contains a header (control data) and a payload (the actual content). MTU size sets the limit on this payload. If a packet exceeds the MTU of any network component along the path, it must be fragmented—decreasing performance.

Several factors influence MTU size:

  • Network technology: Ethernet supports 1,500-byte MTU by default; jumbo frames (9,000 bytes) are common in data centers.
  • Network devices: Switches, NICs, and routers must all support the configured MTU.
  • Path MTU Discovery (PMTUD): Dynamically finds the maximum MTU along a path.
  • Modern Ethernet standards: IEEE 802.3bs (400G) and 802.3ck (800G) support enhanced jumbo frame handling. Ethernet-APL and enhanced Data Center Bridging (DCB) improve MTU management.
  • IPv6: MTU considerations differ from IPv4, with minimum MTU of 1,280 bytes.

Increasing the MTU size can offer some benefits in certain scenarios, including: 

  • Reduced overhead: Larger MTU sizes reduce the ratio of header overhead to data payload, resulting in more efficient data transmission and potentially improved network performance.
  • Throughput optimization: With a larger MTU size, fewer packets are required to transmit a given amount of data. This can lead to increased throughput, especially in high-bandwidth scenarios.

Changing the MTU size can also have downsides:

  • Compatibility issues: Some devices or applications may not handle non-standard MTU sizes well, causing compatibility issues.
  • Path limitations: While larger MTU sizes can be beneficial within a local network, they may encounter limitations when traversing certain paths, such as tunnels or links with smaller MTU size restrictions.

The MTU size itself does not directly affect network speed. It does, however, impact the efficiency of data transmission and can have implications for network performance. By optimizing the MTU size, you can potentially improve network throughput and reduce latency, leading to a better overall user experience.

Changing the MTU size should be done cautiously and only if there is a specific need or optimization goal. In most cases, the default MTU size provided by the network technology and infrastructure is sufficient for regular network usage. However, if you encounter specific issues or have advanced requirements, adjusting the MTU size might be worth considering.

There is no definitive answer to whether a high or low MTU size is better. The optimal MTU size depends on various factors, including the network technology, devices involved, and specific network requirements. In general, a higher MTU size can be beneficial for reducing overhead and optimizing throughput, but it should be compatible with the network infrastructure and paths traversed.

Jumbo Frames and NVMe/TCP MTU: Why Size Matters

Modern storage protocols—particularly NVMe/TCP—benefit tremendously from using jumbo frames. Setting MTU to 9,000 bytes reduces protocol overhead and significantly improves efficiency.

NVMe/TCP MTU sizing: NVMe/TCP benefits significantly from 9,000-byte MTU, reducing CPU overhead by up to 40% compared to 1,500-byte frames.

Storage multipathing impact: Different MTU sizes across paths can cause failover delays. Ensure consistent MTU across all storage network paths.

Storage Protocol  MTU Requirements (Comparison Table)

ProtocolRecommended MTUImpact of Incorrect MTU
NVMe/TCP9,000 bytes40% CPU overhead increase with 1,500 bytes
iSCSI9,000 bytesIncreased latency and reduced IOPS
SMB 3.x9,000 bytesPoor large file transfer performance
NFS v4.1+9,000 bytesReduced throughput for sequential workloads
FCN/A (frame-based)Consistent 2KB frames; not affected by MTU

Pure Storage FlashArray MTU Recommendations

FlashArray™ systems are engineered to take full advantage of optimal MTU configurations:

  • Management networks: 1,500 bytes (standard Ethernet MTU)
  • iSCSI data networks: 9,000 bytes (jumbo frames)
  • NVMe/TCP networks: 9,000 bytes (required for optimal performance)
  • Replication networks: 9,000 bytes (reduces bandwidth overhead)

How to Change MTU Size

Check current MTU:

Change MTU:

Test with ping:

Linux and container examples:

Storage-specific testing:

Modern Data Center Considerations

Network virtualization: VMware NSX-T and Cisco ACI overlay networks impact MTU. MTU must accommodate encapsulation overhead (typically +50 bytes), requiring careful planning.

Kubernetes Container Network Interface (CNI): Different CNIs (Calico, Flannel, etc.) may enforce their own MTU limits. Container orchestrators may override host MTU settings. Configure storage classes with appropriate MTU parameters.

Multi-cloud storage: Stretching storage networks across clouds introduces limitations. Cloud provider MTU limits (typically 1,500 bytes) can bottleneck hybrid storage performance.

Edge computing: Edge environments may lack jumbo frame support. Reduced MTU settings (1,500 bytes) are often necessary to accommodate constrained networks.

Pure Storage Integration Updates

Replace legacy SAN guidance with modern Pure Storage MTU practices:

Pure Storage MTU Optimization Guide (2025):

  • FlashArray//C™: Automatic MTU detection with NVMe/TCP
  • FlashArray//X™: Supports dynamic MTU adjustment via Pure1® recommendations
  • FlashBlade®: Object storage benefits from 9,000-byte MTU for large file transfers
  • Pure Storage Cloud™: Inherits cloud provider MTU limits

Conclusion

MTU optimization is critical for modern storage performance, especially with NVMe/TCP becoming the standard protocol for all-flash arrays. While 1,500-byte frames work for basic connectivity, 9,000-byte jumbo frames are essential for achieving optimal storage performance and reducing CPU overhead.

Key takeaways for storage administrators:

  • Use 9,000-byte MTU for all storage data networks (iSCSI, NVMe/TCP, SMB)
  • Maintain 1,500-byte MTU for management and out-of-band networks
  • Test MTU changes thoroughly across all storage paths and multipathing configurations
  • Consider container and cloud limitations when designing hybrid storage architectures 

Pure Storage arrays automatically optimize for your network’s MTU configuration, but proper network design amplifies these benefits. FlashArray//C and FlashArray//X provide real-time MTU recommendations through Pure1 analytics, while FlashBlade scales object storage performance linearly with properly configured jumbo frames.

Ready to optimize your storage network? Contact Pure Advanced Services for a comprehensive storage network assessment and MTU optimization strategy tailored to your infrastructure.

FAQ

The maximum transmission unit (MTU) is the largest packet size (in bytes) that can be sent over a network interface without needing to be fragmented. It includes the payload plus protocol headers up to, but not including, the Layer 2 (Ethernet) frame header.

MTU affects both throughput and efficiency. Larger MTUs reduce protocol overhead and CPU load by sending more data per packet, which is especially important for high-bandwidth or storage-heavy traffic. However, if MTU is too large for any device along the path, packets can be dropped or fragmented, hurting performance.

On most Ethernet networks, the default MTU is 1500 bytes. Many data centers and storage networks also use jumbo frames (commonly 9000 bytes) on dedicated segments to improve throughput for large data transfers.

To calculate effective MTU, start with the maximum frame size supported by the network segment, then subtract the overhead for headers (for example, IP and TCP/UDP headers, VLAN tags, and any encapsulation such as VXLAN). The result is the maximum payload size you can safely send without fragmentation.

MTU is the maximum layer 3 packet size, while maximum segment size (MSS) is the largest TCP payload that can fit inside that MTU after accounting for IP and TCP headers. MSS is typically MTU – 40 bytes (20 for IP + 20 for TCP) on IPv4 without options.

Mismatched or overly large MTUs can lead to dropped packets, retransmissions, and poor application performance. In some cases, you’ll see intermittent connectivity issues that are hard to diagnose because small packets succeed while larger ones fail.

Use a consistent MTU across all devices on that path, test end-to-end with tools that send large packets, and favor jumbo frames on dedicated storage or data networks where every hop is under your control. For mixed or internet-facing traffic, staying closer to the default 1500-byte MTU is usually safer.

You can use tools like ping with a “don’t fragment” flag and a gradually increasing packet size to discover the largest packet that traverses the path without fragmentation. Many operating systems and network devices also expose interface MTU via CLI or management GUIs.