full socket buffers

[Mirek Lauš]

On Wed, Feb 9, 2011 at 3:42 PM, Glenn Satchell
<glenn.satchell at uniq.com.au> wrote:
> On 02/10/11 00:40, Mirek Lauš wrote:
>>
>> On Wed, Feb 9, 2011 at 2:38 PM, David Forrest<drf at maplepark.com>  wrote:
>>>
>>> On Wed, 9 Feb 2011, Mirek Lauš wrote:
>>>
>>>> According to this:
>>>> http://www.29west.com/docs/THPM/udp-buffer-sizing.html
>>>> the kern.ipc.maxsockbuf is the right oid. The problem only occurs after
>>>> few days
>>>> of dhcpd process running then it begins do drop UDP randomly forcing us
>>>> to
>>>> restart dhcpd process. Then everything goes well for few days again. Now
>>>> I really don't know where to look for help.
>>>>
>>>> Regards,
>>>> Miroslav
>>>
>>> Perhaps a cron job restarting dhcpd during an inactive time period every
>>> couple days would stop the problem from affecting your users.
>>
>> Yes. But I don't like this solution because it is not a solution. It
>> is a workaround.
>> We would not be able to find the real cause of the issue.
>>
>> Regards,Miroslav
>
> Is there any pattern, such as time of day when it occurs? Any other heavy
> network traffic at the time? Perhaps this is a symptom of some other issue?
> Have you confirmed with netstat -s that the number of packets dropped due to
> full socket buffers increases when the problem is apparent?
>
> How about increasing the buffer size a bit further? 256k is pretty small by
> today's standards. In an earlier post you mentioned you tried 1MB, so
> perhaps try 2MB, then 4, maybe 8MB to see if that is sufficient.
>
> The web page you refer to seems to explain the settings quite well.

No. It is completely random. When the issue occur, the netstat -s -p
udp show continual
loss of UDP packets due to full socket buffers.

I already stated that the kern.ipc.maxsockbuf is set to 8MB and
net.inet.udp.recvspace is set to 1MB. How can I monitor buffer usage?

-m