Re: [PATCH 0/3] Improvements in packet timestamps support

["Nélio Laranjeiro" <nelio.laranjeiro@6wind.com>]

On Thu, Oct 13, 2016 at 02:35:05PM +0000, Oleg Kuporosov wrote:
> 
> Hello DPDK Developers,
> 
> Financial Services Industry which is pretty eager for several DPDK
> features especially low latency while high throughput. The major issue
> so far for increasing DPDK adoption there is requirement for several
> applications (capturers, some trading algorithms) to support of accurate
> timestamping. The requirement mostly came from regulatory and customers
> need strictly follow it.
> 
> Current state of timestamping support in DPDK looks pretty good:
>  - there is API to enable/disable timestamping acquisition by 
>    rte_eth_timesync_enable/disable
>  - get timestamps itself by timesync_read_rx/tx_timestamp
>  - and even implementation of NTP IEEE 1588 for time synchronization
>    by rte_eth_timesync_adjust_read/write_time APIs.
>    
> But it misses the most important feature there – embedding timestamp
> in rte_mbuf aligning it with packet.
> 
> We would like to change this to increase DPDK adoption for several new
> DPDK-based applications for FSI segment. It also might be very
> applicable for several RT media and debugging purposes of network
> flows/streams in other segments like HPC.
> 
> There are several thoughts how to improve there:
>  - include uint64_t timestamp field into rte_mbuf with minimal impact
>    to throughput/latency. Keep it just simple uint64_t in ns (more than
>    580 years) would be enough for immediate needs while using full
>    struct timespec with twice bigger size would have much stronger
>    performance impact as missed cacheline0. It is possible as there is
>    6-bytes gap in 1st cacheline (fast path) and moving uint16_t
>    vlan_tci_outer field to 2nd cacheline. 
>  - such move will only impact for pretty rare usable VLAN RX stripping
>    mode for outer TCI (it used only for one NIC i40e from the whole
>    set and keep minimal performance impact for timestamps.  
>  - PMD can fill the field in their callback completion routines
>    depending on enabling this feature in runtime.
> 
> We evaluated other possible options but looks it will have even worse
> performance impact.
> 
> 
> Oleg Kuporosov (3):
>   mbuf: embedding timestamp into the packet
>   app/testpmd: enabled control for packet timestamps
>   net/mlx5: implementation of Rx packet timestamping support
> 
>  app/test-pmd/cmdline.c                      | 122 +++++++++++++++
>  app/test-pmd/parameters.c                   |   4 +
>  app/test-pmd/testpmd.c                      |   5 +
>  app/test-pmd/testpmd.h                      |   1 +
>  doc/guides/testpmd_app_ug/run_app.rst       |   4 +
>  doc/guides/testpmd_app_ug/testpmd_funcs.rst |   7 +
>  drivers/net/mlx5/mlx5.c                     |   7 +-
>  drivers/net/mlx5/mlx5.h                     |  10 +-
>  drivers/net/mlx5/mlx5_defs.h                |   4 +
>  drivers/net/mlx5/mlx5_ethdev.c              | 222 +++++++++++++++++++++++++++-
>  drivers/net/mlx5/mlx5_rxq.c                 |   2 +
>  drivers/net/mlx5/mlx5_rxtx.c                |  19 ++-
>  drivers/net/mlx5/mlx5_rxtx.h                |   7 +-
>  drivers/net/mlx5/mlx5_time.h                |  53 +++++++
>  drivers/net/mlx5/mlx5_trigger.c             |   1 +
>  lib/librte_eal/common/include/rte_time.h    |  45 ++++++
>  lib/librte_mbuf/rte_mbuf.h                  |   6 +-
>  17 files changed, 507 insertions(+), 12 deletions(-)
>  create mode 100644 drivers/net/mlx5/mlx5_time.h
> 
> Thanks,
> Oleg.
> -- 
> 1.8.3.1

Reviewed-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>

-- 
Nélio Laranjeiro
6WIND