Hello,

additional information:

- when the TX ring size is set to 24, we experience no system freeze, but
the connectivity is severely impaired:

PING 143.225.28.254 (143.225.28.254) 56(84) bytes of data.
64 bytes from 143.225.28.254: icmp_seq=5 ttl=255 time=0.332 ms
^C
--- 143.225.28.254 ping statistics ---
18 packets transmitted, 1 received, 94% packet loss, time 17229ms
rtt min/avg/max/mdev = 0.332/0.332/0.332/0.000 ms


- when the TX ring size is set to 32, the percentage of ping packet loss is
0% (throughput not investigated yet).

Any idea how to explain the different behaviors when the TX ring size is
set to 16, 24 and 32?

Thanks a lot in advance.

Bests,
Stefano


On Wed, Nov 16, 2016 at 11:08 AM, Stefano Avallone <stava...@gmail.com>
wrote:

> Hello,
>
> we want to make experiments with smaller TX ring sizes to evaluate how
> much the throughput decreases. We are using two distinct NICs on two
> different PCs:
>
> Ethernet controller: Intel Corporation Ethernet Connection (2) I219-LM
> (rev 31)
>
> and
>
> Ethernet controller: Intel Corporation Ethernet Connection I217-LM (rev
> 04)
>
> both using the e1000e kernel module (kernel 4.8.7, module version 3.2.6)
>
> We recompiled the e1000e module with the following patch:
>
> --- linux-4.8/drivers/net/ethernet/intel/e1000e/e1000.h.orig
>  2016-11-15 15:42:18.181412740 +0100
> +++ linux-4.8/drivers/net/ethernet/intel/e1000e/e1000.h 2016-11-15
> 15:42:55.961607611 +0100
> @@ -64,7 +64,7 @@
>  /* Tx/Rx descriptor defines */
>  #define E1000_DEFAULT_TXD              256
>  #define E1000_MAX_TXD                  4096
> -#define E1000_MIN_TXD                  64
> +#define E1000_MIN_TXD                  4
>
>  #define E1000_DEFAULT_RXD              256
>  #define E1000_MAX_RXD                  4096
>
> Then, we set the TX ring size to 16 with ethtool:
>
> sudo ethtool -G <iface> tx 16
>
> The change was applied (as confirmed by ethtool -g) but the whole system
> freezed a few seconds later. This happens every time we set the tx ring
> size to 16. We cannot ping the machine nor ssh into it, we can only force
> shutdown. Looking at the journal log (systemd) there is no sign of the
> freeze.
>
> We would like to know if there are known issues that prevent setting the
> TX ring size below a certain threshold. Also, we would appreciate it very
> much if someone could advise what to do to investigate/solve the issue we
> are facing.
>
> Thanks a lot in advance for any help.
>
> Bests,
> Stefano
>
>
------------------------------------------------------------------------------
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel&#174; Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to