Re: [patch] voluntary-preempt-2.6.9-rc1-bk4-R1

[Ingo Molnar]

* Mark_H_Johnson@xxxxxxxxxxxx <Mark_H_Johnson@xxxxxxxxxxxx> wrote:

> Results from this morning's test with -R1 and some follow up on
> related messages.

lt002.v3k1/lt.10 is particularly interesting:

 00000001 0.000ms (+0.000ms): spin_lock (rtl8139_poll)
 00000001 0.000ms (+0.070ms): spin_lock (<00000000>)
 00000001 0.070ms (+0.070ms): rtl8139_rx (rtl8139_poll)
 00000001 0.140ms (+0.070ms): alloc_skb (rtl8139_rx)
 00000001 0.210ms (+0.070ms): kmem_cache_alloc (alloc_skb)
 00000001 0.280ms (+0.073ms): __kmalloc (alloc_skb)
 00000001 0.354ms (+0.139ms): eth_type_trans (rtl8139_rx)
 00000001 0.493ms (+0.076ms): netif_receive_skb (rtl8139_rx)
 00000002 0.570ms (+0.001ms): packet_rcv_spkt (netif_receive_skb)

this too shows the CPU in 'slow motion' in a codepath that normally
executes 10 times faster than this on a 100 MHz Pentium Classic ...

another interesting thing is that the unit of delay seems to be around
70 usecs. As if under certain circumstances every main memory access
created a 70 usecs hit. (a cachemiss perhaps?) The eth_type_trans entry
perhaps generated 2 main memory accesses (2 cachemisses?) or so.

	Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/