I have been trying to see if there is any improvement with setting the tcp recv highwater marks and I have not seen any improvement. Increasing the water marks actually worsens preformance in our enviroment, this maybe due to the buffer size that cvs may use. I have also played with the ipg settings and still no improvement. Below are the settings that are in use on our 2.8 machines:
/dev/hme: transceiver_inuse 0 link_status 1 link_speed 1 link_mode 1 ipg1 8 ipg2 4 use_int_xcvr 0 pace_size 0 adv_autoneg_cap 0 adv_100T4_cap 0 adv_100fdx_cap 1 adv_100hdx_cap 0 adv_10fdx_cap 0 adv_10hdx_cap 0 autoneg_cap 1 100T4_cap 0 100fdx_cap 1 100hdx_cap 1 10fdx_cap 1 10hdx_cap 1 lp_autoneg_cap 0 lp_100T4_cap 0 lp_100fdx_cap 0 lp_100hdx_cap 0 lp_10fdx_cap 0 lp_10hdx_cap 0 instance 0 lance_mode 1 ipg0 16
/dev/tcp: tcp_time_wait_interval 120000 tcp_conn_req_max_q 128 tcp_conn_req_max_q0 4096 tcp_conn_req_min 1 tcp_conn_grace_period 0 tcp_cwnd_max 1048576 tcp_debug 0 tcp_smallest_nonpriv_port 1024 tcp_ip_abort_cinterval 180000 tcp_ip_abort_linterval 180000 tcp_ip_abort_interval 480000 tcp_ip_notify_cinterval 10000 tcp_ip_notify_interval 10000 tcp_ipv4_ttl 64 tcp_keepalive_interval 7200000 tcp_maxpsz_multiplier 2 tcp_mss_def_ipv4 536 tcp_mss_max_ipv4 65495 tcp_mss_min 1 tcp_naglim_def 4095 tcp_rexmit_interval_initial 3000 tcp_rexmit_interval_max 60000 tcp_rexmit_interval_min 400 tcp_deferred_ack_interval 100 tcp_snd_lowat_fraction 0 tcp_sth_rcv_hiwat 0 tcp_sth_rcv_lowat 0 tcp_dupack_fast_retransmit 3 tcp_ignore_path_mtu 0 tcp_rcv_push_wait 16384 tcp_smallest_anon_port 32768 tcp_largest_anon_port 65535 tcp_xmit_hiwat 16384 tcp_xmit_lowat 4096 tcp_recv_hiwat 24576 tcp_recv_hiwat_minmss 4 tcp_fin_wait_2_flush_interval 675000 tcp_co_min 64 tcp_max_buf 1048576 tcp_strong_iss 2 tcp_rtt_updates 20 tcp_wscale_always 0 tcp_tstamp_always 0 tcp_tstamp_if_wscale 1 tcp_rexmit_interval_extra 0 tcp_deferred_acks_max 8 tcp_slow_start_after_idle 4 tcp_slow_start_initial 2 tcp_co_timer_interval 20 tcp_sack_permitted 2 tcp_trace 0 tcp_compression_enabled 1 tcp_ipv6_hoplimit 60 tcp_mss_def_ipv6 1220 tcp_mss_max_ipv6 65475 tcp_rev_src_routes 0 tcp_wroff_xtra 32
edwardv-
MCope@isisph.com wrote:
Edward,
What do your Solaris systems look like? There are a number of things that can be done to get Sun systems, and especially their NICs, to coexist with Filers. I'm currently having a lot of fun getting some temperamental SBUS Gigabit NICs to perform up to spec (same problem - writes take FOREVER)
Michael
edward valencia edjv@corp.earthlink.net To: Sent by: "toasters@mathworks.com" owner-toasters@mathworks.com toasters@mathworks.com cc: 02/13/02 05:14 AM Subject: Slow writes
Hi all,
I hope that someone has gone through this as I have been trying to figure this out and its making me go nuts. On our solaris(2.8, 2.6) we usually do small file writes, our users do cvs checkouts to thier home directories and compile software that they are working on in that directory also. The problems is not the reads but the writes, I have changed the client machines to nfs3/tcp with rsize and wsize or 32k. The avg time todo a cvs checkout has been about 24 seconds, if this is done to local disk we are seeing the avg. of 4 seconds. The size of the module they are checking out is 4mb, this is what I have been doing tests with, but other cvs modules are also slow while being written to the nfs mount home directory. The filers also have the nfs.tcp.xfersize of 32k. Here are the configs of the filer:
F760 NetApp Release 6.0.1R3: Fri Feb 16 02:42:49 PST 2001
e0: flags=8042<BROADCAST,RUNNING,MULTICAST> mtu 1500 inet 10.174.32.67 netmask 0xffffff80 broadcast 10.174.32.127 ether 00:a0:98:00:56:a3 (auto-100tx-down) e4: flags=900042<BROADCAST,RUNNING,TCPCKSUM> mtu 1500 inet 10.174.32.126 netmask 0xffffff80 broadcast 10.174.32.127 ether 00:60:cf:20:31:fb (1000fx) e6: flags=848043<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 inet 10.174.32.7 netmask 0xffffff80 broadcast 10.174.32.127 ether 00:90:27:f4:5b:0e (auto-1000sx-fd-up) flowcontrol full lo: flags=948049<UP,LOOPBACK,RUNNING,MULTICAST,TCPCKSUM> mtu 9188 inet 127.0.0.1 netmask 0xff000000 broadcast 127.0.0.1
We currently only use e6.
nfs.mount_rootonly on nfs.per_client_stats.enable off nfs.tcp.enable on nfs.tcp.xfersize 32768 nfs.udp.xfersize 8192 nfs.v2.df_2gb_lim off nfs.v3.enable on nfs.webnfs.enable off nfs.webnfs.rootdir XXX nfs.webnfs.rootdir.set off
I hope you all can help.
cheers,
edwardv-