On Tue, May 11, 2010 at 01:05:47AM +0300, Andreas Gerdd wrote: > > On Mon, May 10, 2010 at 03:34, Daniel Melameth <[email protected]> > wrote: > > I'll concur--that's pretty slow. Have you tried increasing > > net.inet.tcp.recvspace and/or net.inet.tcp.sendspace > > Increasing TCP send,recv and UDP send,recv dramatically improved the speed > from > 80 KB/s to 1.12M/s. God.. What a difference! > > from the default OpenBSD 4.6 values to: > net.inet.tcp.sendspace=262144 > net.inet.tcp.recvspace=262144 > net.inet.udp.recvspace=262144 > net.inet.udp.sendspace=262144 > Would those high values make the server vulnerable to ddos attacks? >
Yes. First of all you increased the tcp sendspace by 16 so 16 times more memory will be consumed for each socket. So in the end you can use 16 times less concurrent sockets. Watch out for hitting the kern.maxclusters limit. Second the tcp recvspace should normaly not cause any troubles since as long as userland is responding that socketbuffer will stay almost empty. Third bumping udp send space makes no sense at all and the recv space should only be touched if you expect massive burst of huge udp traffic. > I still don't feel comfortable with that bge0 card. Heh 80 KB/s.. > I'll request an Intel NIC from the company. > At 10Mbps. fxp(4), em(4) is overkill but a good option as well. > > On Mon, May 10, 2010 at 09:08, Tomas Bodzar <[email protected]> > wrote: > > If you will take a look trough archives you will find that best option > > is to use Intel cards because of quality of HW. > > As you suggested, what model Intel NIC would be the best possible choice for > OpenBSD? > > > On Mon, May 10, 2010 at 18:46, James Records <[email protected]> > wrote: > > Is this only with http? If you transfer a file using scp is it any > faster? > > Tried both HTTP/FTP/SCP. Same speed. The SFTP/SCP speed was even around some > few bytes! > -- :wq Claudio
