Re: [PATCH 02/20] ipvs: reorganize tot_stats

To: Eric Dumazet <eric.dumazet@xxxxxxxxx>
Subject: Re: [PATCH 02/20] ipvs: reorganize tot_stats
Cc: Simon Horman <horms@xxxxxxxxxxxx>, netdev@xxxxxxxxxxxxxxx, netfilter-devel@xxxxxxxxxxxxxxx, netfilter@xxxxxxxxxxxxxxx, lvs-devel@xxxxxxxxxxxxxxx, Hans Schillstrom <hans@xxxxxxxxxxxxxxx>
From: Julian Anastasov <ja@xxxxxx>
Date: Mon, 14 Mar 2011 11:21:21 +0200 (EET)


On Mon, 14 Mar 2011, Eric Dumazet wrote:

        Also, make sure ip_vs_stats_percpu_show reads properly
the u64 stats from user context.

Signed-off-by: Julian Anastasov <ja@xxxxxx>
Signed-off-by: Simon Horman <horms@xxxxxxxxxxxx>

Reviewed-by: Eric Dumazet <eric.dumazet@xxxxxxxxx>

One comment : per cpu stats are really good to avoid cache misses, but I
see tot_stats is included in struct netns_ipvs right after "conn_count",
possibly sharing a hot cache line ?

        conn_count and tcp_app_lock, udp_app_lock, sctp_app_lock
are used when packet creates new connection, probably on the
same time depending on the protocol in packet. May be conn_count
better to be before tcp_apps, far away from some sysctl vars
that are mostly read?

        For tot_stats: cpustats ptr is always read,
other fields in tot_stats are updated from the 2-second
timer. May be it is better just to move tot_stats before
est_list, it is written at the same time?


Julian Anastasov <ja@xxxxxx>
To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at

<Prev in Thread] Current Thread [Next in Thread>