Problem: ixgbe 0000:d8:00.1: swiotlb buffer is full

SUSE Linux Enterprise Server SLES Networking
1

We use SLES15 SP1 as XEN Virtual Maschine Host.

Jumboframes are turned on. Network degrades on 10GBit Intel Ethernet Card.

Here are the Kernelmessages:

Jan 15 10:31:08 atlas kernel: swiotlb_tbl_map_single: 46 callbacks suppressed
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:08 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:36 atlas kernel: swiotlb_tbl_map_single: 15 callbacks suppressed
Jan 15 10:31:36 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:36 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:37 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:37 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:37 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)
Jan 15 10:31:37 atlas kernel: ixgbe 0000:d8:00.1: swiotlb buffer is full (sz: 8192 bytes)

What can I do to increase the buffers while running?

2

This is a self made issue. Jumbo frames don’t bring any performance gain with a good network card. Good network card has enough hardware acceleration on board (offloading). Check:

# ethtool --show-offload eth0

Switch off jumbo frames and tune your network settings:
https://fasterdata.es.net/host-tuning/linux/100g-tuning/