On Fri, 25 Aug 2006 10:32:38 +0400 Evgeniy Polyakov <[EMAIL PROTECTED]> wrote:
> On Thu, Aug 24, 2006 at 11:20:24PM -0700, Andrew Morton ([EMAIL PROTECTED]) > wrote: > > On Fri, 25 Aug 2006 09:48:15 +0400 > > Evgeniy Polyakov <[EMAIL PROTECTED]> wrote: > > > > > kmalloc is really slow actually - it always shows somewhere on top > > > in profiles and brings noticeble overhead > > > > It shouldn't. Please describe the workload and send the profiles. > > epoll based trivial server (accept + sendfile for the same file, about > 4k), httperf with big amount of simulateneous connections. 3c59x NIC > (with e1000 there were no ioreads and netif_rx). > __alloc_skb calls kmem_cache_alloc() and ___kmalloc(). > > 16158 1.3681 ioread16 > 8073 0.6835 ioread32 > 3485 0.2951 irq_entries_start > 3018 0.2555 _spin_lock > 2103 0.1781 tcp_v4_rcv > 1503 0.1273 sysenter_past_esp > 1492 0.1263 netif_rx > 1459 0.1235 skb_copy_bits > 1422 0.1204 _spin_lock_irqsave > 1145 0.0969 ip_route_input > 983 0.0832 kmem_cache_free > 964 0.0816 __alloc_skb > 926 0.0784 common_interrupt > 891 0.0754 __do_IRQ > 846 0.0716 _read_lock > 826 0.0699 __netif_rx_schedule > 806 0.0682 __kmalloc > 767 0.0649 do_tcp_sendpages > 747 0.0632 __copy_to_user_ll > 744 0.0630 pskb_expand_head > That doesn't look too bad. What's that as a percentage of total user+system time? - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html