Re: [BUG] 2.2.19 -> 80% Packet Loss

Scott Laird (laird@internap.com)
Thu, 14 Jun 2001 16:44:34 -0700 (PDT)


Odds are it's a raw socket receive buffer issue. Stock pings only ask for
a ~96k socket buffer, which means that they can only hold one ~64k packet
at a time. So, if you're ever slow grabbing packets out of the buffer,
you're going to drop traffic.

You can fix this by upping the socket buffer that ping asks for (look for
setsockopt( ... SO_RCVBUF ...)) and then tuning the kernel to allow larger
socket buffers. The file to fiddle with is /proc/sys/net/core/rmem_max.

That doesn't really answer why you'd want to fling that many 64k-ish ping
packets around, though.

Scott

On Thu, 14 Jun 2001 chuckw@altaserv.net wrote:

>
> 1. When pinging a machine using kernel 2.2.19 I consistently get an 80%
> packet loss when doing a ping -f with a packet size of 64590 or higher.
>
>
> 2. A "ping -f -s 64589" to a machine running kernel 2.2.19 results in 0%
> packet loss. By incrementing the packetsize by one "ping -f -s 64590" or
> higher, I consistently see 80% packet loss. ifconfig on the receiving
> machine shows no anomolies.
>
>
> 3. 2.2.19, ping, flood, 64589, 64590, 80%, packet, loss
>
>
> 4. Linux version 2.2.19-7.0.1 (root@porky.devel.redhat.com) (gcc version
> egcs-2.91.66 19990314/Linux (egcs-1.1.2 release)) #1 Tue Apr 10 00:55:03
> EDT 2001
>
>
> 5. There was no oops associated with this.
>
>
> 6. >>EOF
> /bin/ping -f -s 64589
> /bin/ping -f -s 64590
> EOF
>
>
> 7.1 /usr/src/linux-2.2.19/scripts/ver_linux
> Linux orchid 2.2.19-7.0.1 #1 Tue Apr 10 00:55:03 EDT 2001 i686 unknown
>
> Gnu C 2.96
> Gnu make 3.79.1
> binutils 2.10.0.18
> util-linux 2.10r
> modutils 2.3.21
> e2fsprogs 1.18
> Linux C Library > libc.2.2
> Dynamic linker (ldd) 2.2
> Procps 2.0.7
> Net-tools 1.56
> Console-tools 0.3.3
> Sh-utils 2.0
> Modules Loaded autofs nfsd lockd sunrpc 3c59x agpgart usb-uhci
> usbcore
>
>
> 7.2 cat /proc/cpuinfo
> processor : 0
> vendor_id : GenuineIntel
> cpu family : 6
> model : 8
> model name : Pentium III (Coppermine)
> stepping : 3
> cpu MHz : 751.725
> cache size : 256 KB
> fdiv_bug : no
> hlt_bug : no
> sep_bug : no
> f00f_bug : no
> coma_bug : no
> fpu : yes
> fpu_exception : yes
> cpuid level : 2
> wp : yes
> flags : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov
> pat pse36 mmx fxsr xmm
> bogomips : 1500.77
>
>
> 7.3 cat /proc/modules
> autofs 9424 3 (autoclean)
> nfsd 182752 8 (autoclean)
> lockd 45264 1 (autoclean) [nfsd]
> sunrpc 61808 1 (autoclean) [nfsd lockd]
> 3c59x 21584 1 (autoclean)
> agpgart 18960 0 (unused)
> usb-uhci 18736 0 (unused)
> usbcore 43120 1 [usb-uhci]
>
>
> 7.4 No SCSI devices
>
>
> 7.5 /sbin/ifconfig (after sending it several pings with 80% packet loss)
>
> eth0 Link encap:Ethernet HWaddr 00:60:97:D7:60:E4
> inet addr:10.0.0.102 Bcast:10.0.0.255 Mask:255.255.255.0
> UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
> RX packets:3972983 errors:0 dropped:0 overruns:0 frame:0
> TX packets:5466442 errors:0 dropped:0 overruns:0 carrier:0
> collisions:0 txqueuelen:100
> Interrupt:12 Base address:0xa800
>
> lo Link encap:Local Loopback
> inet addr:127.0.0.1 Mask:255.0.0.0
> UP LOOPBACK RUNNING MTU:3924 Metric:1
> RX packets:210 errors:0 dropped:0 overruns:0 frame:0
> TX packets:210 errors:0 dropped:0 overruns:0 carrier:0
> collisions:0 txqueuelen:0
>
>
> X. This has only been tested/observed on a 10Mb as well as a newer 100Mb
> LAN. This has also been observed on machines running kernel 2.2.17. Non
> Linux machines (NT and Win98) were tested in the same manner and do not
> show the same symptoms. I reviewed the LKML archives, grep'd in the
> sources and did a string on the kernel binary and was not able to find any
> useful reference to 64590, 64589 or ping losses due to packet size.
>
> --
> Chuck Wolber
> System Administrator
> AltaServ Corporation
> (425)576-1202
> ten.vresatla@wkcuhc
>
> Quidquid latine dictum sit, altum viditur.
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>

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