On 4/10/2019 5:50 PM, Ben Pfaff wrote:
On Wed, Apr 10, 2019 at 05:35:34PM +0100, Ian Stokes wrote:
On 4/10/2019 5:21 PM, Ian Stokes wrote:
On 3/5/2019 4:28 PM, Ilya Maximets wrote:
'vhost_id' is an array of 'PATH_MAX' bytes in the middle of
'netdev_dpdk' structure. That is 4K bytes.
'vhost_id' never used on a hot path and there is no need to keep
it inside the structure memory. Dynamic allocation will allow to
decrease 'struct netdev_dpdk' signigficantly, saving 4KB per ETH
signigficantly -> significantly
port (ETH ports doesn't use 'vhost_id') and almost same value per
vhost ports (real 'vhost_id's, in common case, are much shorter).
We could save the pointer space by making the union with 'devargs'
which is mutually exclusive with 'vhost_id'.
As we're just removing the single 'PADDED_MEMBER', the total
cacheline layout is not affected.
Stats for 'struct netdev_dpdk':
Before: /* size: 4992, cachelines: 78 */
After : /* size: 896, cachelines: 14 */
Thanks Ilya, this looks like a good change, validated without issue.
I've addressed a minor typo in the commit message and comment style
issue below. Other than that it looks good so applied to master.
On a broader note, do you think it's worth capturing changes such as
this for users in documentation? i.e. previously the vhost-id path was
limited to 4096 (however unlikely it was that someone would exceed
this), but it wasn't captured in the OVS documentation from what I can
see. From a usability point is it something to document between the 2.10
and 2.11 release?
Sorry, meant OVS 2.11 to 2.12 above.
I like to document user-visible changes, but I think in this case it is
unlikely to fix any real user problems. (It would be different if we
made this change in response to a user having a problem with a very long
vhost-id path.)
Sure, I don't think a user would be aware of this unless they were
hitting the 4096 limit, in which case I think they would have a
different issue at hand :). It was just an after thought during the
review when I checked if the vhost-user path limit was documented. Not
necessary to be captured now though.
Ian
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
