We have 4 x t2000 and 2 x t1000 sun servers all running solaris 10 sparc edition and when uploading an agent to all these servers I was unable to maintain a connection faster than 6MB. All our OEL 6.3 hosts on the same vlans/network/swithes are able to copy the same files around at 85MB/s.
Our entire network is GB and have been trouble shooting along the way and have got down to two sun hosts on the same switch and vlan to rule out firewalling and other factors.
The switchports on the cisco show no errors and no errors are seen on the solaris servers and training up at the GB speed as configured.
So I'm faily certain there is something on the solaris hosts OS wise that is affecting perfomance that is causing some of our backup jobs to be quite long but I'm not sure at what to do now.
I've seen some online articles relating to tuning and adjusting certain tcp options but unsure as to what would be best.
Has anyone else experienced issues such a this and if so what was done to resolve it, thanks.
this is just a guess, but the T1000 and T2000 are really slow when doing single-thread cpu stuff. When you do scp, a single thread or process on each machine has to do all the encryption and this is most likely what is slowing you down. To analyze this further, look at the output of prstat (or top). If I am right, you will see one scp/ssh process using all cpu time of one of your cores/threads. In a T2000 with 8 cores, 32threads, this will be shown in prstat as 6.7% utilization.
After opening a case with Oracle this was narrowed down to the patch level of the solaris hosts. In the current version of solaris the SCP/SSH isnt being offloading to the on-chip crypto that the T2000 server has causing the copy to be VERY slow. If I use FTP or SCP with a different cipher, arcfour, the copies are MUCH faster.