How should I check MTU settings across sites and storage networks?

November 16, 2026 5 answers
Asked by 4k!clayratchet_819

We're seeing inconsistent latency and speeds between sites, with a mix of firewalls, switches, virtual hosts and physical servers. Some traffic shares a subnet with storage traffic.

Should everything stay at 1500, or is there a useful way to run jumbo frames for storage? Where would you start troubleshooting?

Answers

Rilvex+2h

Start with the path between sites before changing the whole network. A VPN or SD-WAN tunnel adds encapsulation, so a 1500 byte IP packet may no longer fit in a 1500 byte underlay. The usable size depends on the tunnel and the links underneath it.

Map a slow flow from endpoint to endpoint, including the firewalls, tunnel and WAN service. Check the MTU at each point and look for drops or fragmentation. An MTU problem is plausible, but don't assume it explains all the latency and speed variation.

_a79+brisketoggle.5:

Some WAN links have less than 1500 available even before a VPN is added. Include the provider handoff and any PPPoE overhead in the path check.

Reply to this answer
_r61+mapwinder.8

I'd leave ordinary client and VM traffic at 1500 for now. Jumbo frames can be useful for storage, but they're not an automatic performance upgrade.

If testing shows a benefit, give storage a dedicated VLAN or subnet and verify the entire storage path supports the chosen size. That includes host interfaces, virtual switches, physical switch ports and storage interfaces. A switch configured to pass jumbo frames doesn't make hosts send them.

4k!clayratchet_819:

Does mixed MTU on the same subnet always break traffic?

_r38+gearwhistle.6:

Not necessarily, but a jumbo sender can run into trouble reaching a smaller-MTU endpoint or crossing a path that can't carry its packets. Keeping storage separate makes those paths easier to verify.

Reply to this answer
v.392!sparecanvas

Test the actual path with different packet sizes, not just a ping that happens to succeed. For IPv4, use a ping option that sets Don't Fragment and increase the payload until it fails. On Linux, ping -4 -M do -s 1472 tests a 1500 byte IPv4 packet when there are no IP options. Run it against a reachable endpoint across the site link and adjust the size downward.

For IPv6, a 1452 byte ping payload normally makes a 1500 byte packet. Check the ping options for your operating system, and allow the ICMP messages needed for path MTU discovery, including ICMPv6 Packet Too Big. A failed ping alone isn't proof of an MTU problem because filtering can also block the test.

_x46~copperstaple.3:

Test both directions where possible. The paths or filtering rules may differ, so a result in one direction doesn't establish the other.

Reply to this answer
_t48!radarcrank.9

Also check TCP MSS clamping on the tunnel if TCP connections stall or perform badly with larger transfers. It can help TCP peers choose smaller segments, but it won't fix every protocol or replace working path MTU discovery.

Be careful with the Don't Fragment bit too. In IPv4 it affects whether an oversized packet can be fragmented by a router. Drops can look intermittent when only some packets exceed the available MTU.

Reply to this answer
Kelmor-7v

Check what each vendor means by MTU before comparing numbers. Some settings refer to the IP packet, while others describe the Ethernet frame a port can pass. Ethernet headers, tags and tunnel overhead are not all counted the same way.

Document the limits along one affected path first. That will make the settings much easier to compare than trying to make every screen show the same number.

Reply to this answer

Add your answer

Answers are reviewed before publication.

Related Discussions