Hi, I'm running som load tests on three machines, one server and two machines that simulates some 3000+ clients each. Each simulated client uses a single TCP/IP connection to the server which generates about 80 kbps of continuous data. One of my client test programs, a Java program, died unexpected with the following in /var/log/messages. (The machine is a IBM xSeries x335 (2x2.4GHz) running linux 2.6.4.) Apr 7 09:55:58 kernel: java: page allocation failure. order:0, mode:0x20 Apr 7 09:55:58 kernel: Call Trace: Apr 7 09:55:58 kernel: [<c013a79c>] __alloc_pages+0x300/0x345 Apr 7 09:55:58 kernel: [<c013a806>] __get_free_pages+0x25/0x3f Apr 7 09:55:58 kernel: [<c013d259>] cache_grow+0x98/0x299 Apr 7 09:55:58 kernel: [<c013d529>] cache_alloc_refill+0xcf/0x215 Apr 7 09:55:58 kernel: [<c013d93d>] __kmalloc+0x73/0x7a Apr 7 09:55:58 kernel: [<c02a14fd>] alloc_skb+0x47/0xe0 Apr 7 09:55:58 kernel: [<f8a73e85>] tg3_alloc_rx_skb+0x8f/0x127 [tg3] Apr 7 09:55:58 kernel: [<f8a741cc>] tg3_rx+0x21e/0x3c5 [tg3] Apr 7 09:55:58 kernel: [<f8a743f8>] tg3_poll+0x85/0x131 [tg3] Apr 7 09:55:58 kernel: [<c02a6508>] net_rx_action+0x7a/0x103 Apr 7 09:55:58 kernel: [<c01222bb>] do_softirq+0xc3/0xc5 Apr 7 09:55:58 kernel: [<c010b5dd>] do_IRQ+0xf6/0x114 Apr 7 09:55:58 kernel: [<c0109968>] common_interrupt+0x18/0x20 Apr 7 09:55:58 kernel: I interpret this as the tg3 module was unable to save incoming network data to a buffer. Is this something one could expect when putting to high values in the /proc/sys/net/ipv4/tcp_rmem and/or /proc/sys/net/ipv4/tcp_mem Regards, Niclas Gustafsson - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
