| Message ID | 20200125102645.4782-1-steffen.klassert@secunet.com |
|---|---|
| Headers | show |
| Series | Support fraglist GRO/GSO | expand |
From: Steffen Klassert <steffen.klassert@secunet.com> Date: Sat, 25 Jan 2020 11:26:41 +0100 > This patchset adds support to do GRO/GSO by chaining packets > of the same flow at the SKB frag_list pointer. This avoids > the overhead to merge payloads into one big packet, and > on the other end, if GSO is needed it avoids the overhead > of splitting the big packet back to the native form. ... Series applied, thank you Steffen.
This patchset adds support to do GRO/GSO by chaining packets of the same flow at the SKB frag_list pointer. This avoids the overhead to merge payloads into one big packet, and on the other end, if GSO is needed it avoids the overhead of splitting the big packet back to the native form. Patch 1 adds netdev feature flags to enable fraglist GRO, this implements one of the configuration options discussed at netconf 2019. Patch 2 adds a netdev software feature set that defaults to off and assigns the new fraglist GRO feature flag to it. Patch 3 adds the core infrastructure to do fraglist GRO/GSO. Patch 4 enables UDP to use fraglist GRO/GSO if configured. I have only meaningful forwarding performance measurements. I did some tests for the local receive path with netperf and iperf, but in this case the sender that generates the packets is the bottleneck. So the benchmarks are not that meaningful for the receive path. Paolo Abeni did some benchmarks of the local receive path for the RFC v2 version of this pachset, results can be found here: https://www.spinics.net/lists/netdev/msg551158.html I used my IPsec forwarding test setup for the performance measurements: ------------ ------------ -->| router 1 |-------->| router 2 |-- | ------------ ------------ | | | | -------------------- | --------|Spirent Testcenter|<---------- -------------------- net-next (September 7th 2019): Single stream UDP frame size 1460 Bytes: 1.161.000 fps (13.5 Gbps). ---------------------------------------------------------------------- net-next (September 7th 2019) + standard UDP GRO/GSO (not implemented in this patchset): Single stream UDP frame size 1460 Bytes: 1.801.000 fps (21 Gbps). ---------------------------------------------------------------------- net-next (September 7th 2019) + fraglist UDP GRO/GSO: Single stream UDP frame size 1460 Bytes: 2.860.000 fps (33.4 Gbps). ======================================================================= net-next (January 23th 2020): Single stream UDP frame size 1460 Bytes: 919.000 fps (10.73 Gbps). ---------------------------------------------------------------------- net-next (January 23th 2020) + fraglist UDP GRO/GSO: Single stream UDP frame size 1460 Bytes: 2.430.000 fps (28.38 Gbps). ----------------------------------------------------------------------- Changes from RFC v1: - Add IPv6 support. - Split patchset to enable UDP GRO by default before adding fraglist GRO support. - Mark fraglist GRO packets as CHECKSUM_NONE. - Take a refcount on the first segment skb when doing fraglist segmentation. With this we can use the same error handling path as with standard segmentation. Changes from RFC v2: - Add a netdev feature flag to configure listifyed GRO. - Fix UDP GRO enabling for IPv6. - Fix a rcu_read_lock() imbalance. - Fix error path in skb_segment_list(). Changes from RFC v3: - Rename NETIF_F_GRO_LIST to NETIF_F_GRO_FRAGLIST and add NETIF_F_GSO_FRAGLIST. - Move introduction of SKB_GSO_FRAGLIST to patch 2. - Use udpv6_encap_needed_key instead of udp_encap_needed_key in IPv6. - Move some missplaced code from patch 5 to patch 1 where it belongs to. Changes from RFC v4: - Drop the 'UDP: enable GRO by default' patch for now. Standard UDP GRO is not changed with this patchset. - Rebase to net-next current. Changes fom v1 (December 18th): - Do a full __copy_skb_header instead of tryng to find the really needed subset header fields. Thisa can be done later. - Mark all fraglist GRO packets with CHECKSUM_UNNECESSARY. - Rebase to net-next current. Changes fom v2 (January 24th): - Do the CHECKSUM_UNNECESSARY setting from IPv4 for IPv6 too.