On Thu, Jun 14, 2012 at 01:54:33PM -0500, Tristin Davis wrote:
> Upgrading is simply not an option. It all comes down to having the
> engineering staff, money, and downtime available. Unfortunatly, we have
> none of the above right now. I realize we *need* to upgrade, but right
> now, tuning the kernel is the only option.
>
> I would be extremely greatful if anyone out there able to provide the
> requested information. :-)
How many engineers does it take to do an upgrade? How much money? Both
close to 0. Yes it takes some time, but your seem to run something so
vital that you should have a backup system anyway.
Now, of the paremeters you mention I know that net.bpf.bufsize will
effect only if you use bpf (for tcpdump etc) so it will not
have any effect on sftp, and you are better of leaving it. For the others,
well, things have changed too much these years, and just don't know.
-Otto
>
>
> Tristin
>
>
>
>
> On Thu, Jun 14, 2012 at 11:52 AM, Chris Zakelj <c.zak...@ieee.org> wrote:
>
> > 1. Any particular reason your systems have not been updated in at least
> > three years?
> > 2. Kernel defaults are generally chosen for a reason. Unless you have
> > evidence of resource exhaustion, turning knobs generally won't make things
> > better, and could make them worse.
> >
> > On Thu, Jun 14, 2012 at 12:31 PM, Tristin Davis
> > <tristin.co...@gmail.com>wrote:
> >
> >> Hardware Type: Intel
> >> Version: OpenBSD 4.3
> >> Kernel: MP
> >>
> >>
> >> I am currently researching some tweaks to increase our network throughput
> >> on a production server. I have found 3 parameters that I may be interested
> >> in changing our systems. I have researched and searched google for
> >> information and impact for the parameters to with little success. What I
> >> am looking for is an idea of what long term impact of changing these
> >> parameters and an explicit description of what the modification of each
> >> kernel parameter will do. These are production sftp servers with a large
> >> number of connections that are rarely rebooted so it is very important
> >> that
> >> changing these values not cause the system to run out of memory. Can
> >> anyone help with the descriptions and possible reprecussions of each of
> >> these parameters?
> >>
> >>
> >> kern.bufcachepercent=90 # Allow the kernel to use up to 90% of the
> >> RAM for cache (default 10%) <-- Suposally not changable via sysctl
> >> kern.maxclusters=128000 # Cluster allocation limit
> >> net.bpf.bufsize=1048576 # Internal kernel buffer for storing packet
> >> captured packets received from the network
> >>
> >>
> >> Thank you in advance,
> >>
> >> Tristin
