deri at ntop
May 9, 2011, 1:26 AM
Post #2 of 2
(218 views)
Permalink
|
|
Re: flowSlotInfo and L1 VIVT coherence issues
[In reply to]
|
|
Rui
my colleague Alfredo is working at memory alignment, so I think he should speak about that.
Alfredo; what's you opinion?
Luca
On May 9, 2011, at 10:23 AM, Rui wrote:
> hi luca:
>
> regarding this data structure,when I compiled it at a 64bit machine,
> the tot_fwd_notok offset is 64 Bytes
> padding offset is 72 Bytes, I wonder if it is comply to your designed?
>
> I think the pack size is 8 Bytes, which lead to tot_pkts offset is
> 24Bytes,not 20Bytes??
>
>
> typedef struct flowSlotInfo {
> /* first page, managed by kernel */
> u_int16_t version, sample_rate;
> u_int32_t min_num_slots, slot_len, data_len, tot_mem;
> u_int64_t tot_pkts, tot_lost, tot_insert;
> u_int32_t insert_off /* managed by kernel */;
> u_int64_t tot_fwd_ok, tot_fwd_notok;
> /* <-- 64 bytes here, should be enough to avoid some L1 VIVT
> coherence issues (32 ~ 64bytes lines) */
> char padding[128];
> /* <-- 128 bytes here, should be enough to avoid false sharing in
> most L2 (64 ~ 128bytes lines) */
> char k_padding[3904];
> /* <-- 4096 bytes here, to get a page aligned block writable by
> kernel side only */
>
> /* second page, managed by userland */
> u_int64_t tot_read;
> u_int32_t remove_off /* managed by userland */;
> u_int32_t vpfring_guest_flags; /* used by vPFRing */
> char u_padding[4080];
> /* <-- 8192 bytes here, to get a page aligned block writable by
> userland only */
> } FlowSlotInfo;
>
>
> regards,
> rui
> _______________________________________________
> Ntop-misc mailing list
> Ntop-misc [at] listgateway
> http://listgateway.unipi.it/mailman/listinfo/ntop-misc
---
If you can not measure it, you can not improve it - Lord Kelvin
_______________________________________________
Ntop-misc mailing list
Ntop-misc [at] listgateway
http://listgateway.unipi.it/mailman/listinfo/ntop-misc
|
