Last week I'd asked for help on speeding up NFS over gigabit for large file transfers (several sybase databases over 100-500GB each) dumped via NFS to v240s. Thanks to Joe Fletcher and Vitaly Porotikov for useful information. Joe suggested that I just dump my v880's and go to x64. Vitaly suggested jumbo frames. I can't do either at the moment but I can look at their other suggestions: Joe suggested tweaking maxphys set maxphys=1048576 set md:md_maxphys=1048576 which I already had set and he has the following tweaks in a startup script: #!/bin/sh # # # Setup for Giga Bit Ethernet # case "$1" in 'start') /usr/sbin/ndd -set /dev/tcp tcp_xmit_hiwat 65535 /usr/sbin/ndd -set /dev/tcp tcp_recv_hiwat 65535 /usr/sbin/ndd -set /dev/udp udp_xmit_hiwat 65535 /usr/sbin/ndd -set /dev/udp udp_recv_hiwat 65535 /usr/sbin/ndd -set /dev/tcp tcp_max_buf 262144 /usr/sbin/ndd -set /dev/tcp tcp_deferred_ack_interval 100 /usr/sbin/ndd -set /dev/tcp tcp_dupack_fast_retransmit 2 /usr/sbin/ndd -set /dev/tcp tcp_slow_start_initial 4 echo gbit_ndd done ;; *) echo "Usage: /etc/init.d/gbit_ndd { start | stop }" ;; esac exit 0 In his testing (he sent me the results) he get's full speed gig-e (100MB/sec) over the wire. ------ Vitaly sent the following tips addressing NFS tuning and TCP tuning: 2. adjust nfs kernel: - nfs3_nra -> 16~32 - nfs3_bsize -> 65536 and hence you can increase rsize (nfs mount option) -> 65536 - nfs3_max_transfer_size -> 65536 - clnt_max_conns -> 16~32 - sq_max_size -> 25~100 use 'mdb -kw' to change on the fly 3. adjust tcp parameters: tcp_max_buf -> 1~4MB tcp_cwnd_max->1~4M tcp_xmit_hiwat -> 131072~1M tcp_recv_hiwat -> 131072~1M tcp_wscale_always -> 1 use ndd to change 4. use 'nocto', 'noac' option for nfs mount Dean Ross-Smith _______________________________________________ sunmanagers mailing list sunmanagers@sunmanagers.org http://www.sunmanagers.org/mailman/listinfo/sunmanagersReceived on Thu Dec 6 10:23:36 2007
This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:44:07 EST