Hi ,
I need you help with some networking related Linux/kernel issues.
1) I do have a question related to setsockopt() too:
I am using the following settings:
/proc/sys/net/core/wmem_default 65535
/proc/sys/net/core/rmem_default 65535
/proc/sys/net/core/wmem_max 65535
/proc/sys/net/core/rmem_max 65535
and I am running an application where I use setsockopt()
to set socket buffer to 80K for example.
I check the setting by using getsockopt and I
am getting 128K instead of 80K.
I looked into the implementation and here is the code:
filename: /usr/src/linux/net/core/sock.c
function: int sock_setsockopt(struct socket *sock, int level, int optname,
char *optval, int optlen)
{
....
case SO_SNDBUF:
/* Don't error on this BSD doesn't and if you think
about it this is right. Otherwise apps have to
play 'guess the biggest size' games. RCVBUF/SNDBUF
are treated in BSD as hints */
if (val > sysctl_wmem_max)
val = sysctl_wmem_max;
sk->sndbuf = max(val*2,2048);
/*
* Wake up sending tasks if we
* upped the value.
*/
sk->write_space(sk);
break;
case SO_RCVBUF:
/* Don't error on this BSD doesn't and if you think
about it this is right. Otherwise apps have to
play 'guess the biggest size' games. RCVBUF/SNDBUF
are treated in BSD as hints */
if (val > sysctl_rmem_max)
val = sysctl_rmem_max;
/* FIXME: is this lower bound the right one? */
sk->rcvbuf = max(val*2,256);
break;
.....
}
QUESTION: why the code is setting sk->sndbuf = max(val*2,2048);
sk->rcvbuf = max(val*2,256);
2*val instead of val ?
Is there any additional kernel space required ?
What will be the Linux equivalent of the BSD mbufs used in net
code ?
In BSD you can control the mbufs settings by manipulating certain
system params and you can get more mbufs for the sockets, e.g.
netstat -m
1 mbufs in use:
1 mbufs allocated to packet headers
0/46 mapped pages in use
124 Kbytes allocated to network (0% in use)
0 requests for memory denied
0 requests for memory delayed
0 calls to protocol drain routines
Is there something similar under Linux ?
Thanks.
2) On a related issue, I do find very useful the BSD-style for netstat -s -p udp
where you can actually say where the errors are coming from , e.g.
udp:
242341 datagrams received
0 with incomplete header
2 with bad data length field
6 with bad checksum
693 dropped due to no socket
198603 broadcast/multicast datagrams dropped due to no socket
531 dropped due to full socket buffers
42506 delivered
9292 PCB hash misses
2738326 datagrams output
In linux we have:
Udp:
122 packets received
17 packets to unknown port received.
0 packet receive errors
141 packets sent
QUESTION: Does anybody know the system calls to determine
the errors "dropped due to full socket buffer ?
Thanks.
3) On another related issue:
I did some tests with bursty traffic using an SMC eth card and I got lots
of transmission errors when I used the default setting in ifconfig TX len
= 100. I had to set it to 1000 to get rid of my errors.
QUESTION: Does anybody know how can I access a similar parameter for RX
queue len ? Since now the bursts are getting dropped at RX side.
Here is the ifconfig of a eth card; the man page of ifconfig let me set
the txqueuelen but no reference to any rxqueuelen.
eth0 Link encap:Ethernet HWaddr 00:E0:18:A8:56:70
inet addr:172.31.17.202 Bcast:172.31.17.255 Mask:255.255.255.0
UP BROADCAST RUNNING MTU:1500 Metric:1
RX packets:86326 errors:1 dropped:0 overruns:0 frame:1
TX packets:582 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
Interrupt:9 Base address:0xb800
Is this a issue with the NIC card or is a kernel setting ?
Thanks,
Marian.
On Tue, 2 May 2000, Glynn Clements wrote:
>
> Asad Khan Awan wrote:
>
> > I wanted to findout that if i keep sending data to a tcp/ip scoket and on
> > the receiving side I dont call the receive (or the read) function then what
> > happnes when the buffer for the socket fills up. I'm talking about the
> > buffer handeled by the kernel, where data is stored till read function is
> > called. can i set the size of this buffer using some function.
>
> int size = ???;
> setsockopt(sock_fd, SOL_SOCKET, SO_SNDBUF, &size, sizeof(size));
> and
> setsockopt(sock_fd, SOL_SOCKET, SO_RCVBUF, &size, sizeof(size));
>
> --
> Glynn Clements <[EMAIL PROTECTED]>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-net" in
> the body of a message to [EMAIL PROTECTED]
>
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
