Re: low gigE transfer rate
From: Eric Gross (ecg5_at_po.cwru.edu)
Date: 01/19/04
- Next message: Darren Dunham: "Re: how to increase "NGROUPS_MAX (supplementary Groups)""
- Previous message: Tony Curtis: "Re: pkgadd problems"
- In reply to: Vahid: "low gigE transfer rate"
- Next in thread: Rick Jones: "Re: low gigE transfer rate"
- Reply: Rick Jones: "Re: low gigE transfer rate"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Mon, 19 Jan 2004 15:34:00 -0500
Are those two machines on the GigE switch always going to be the only
machines on that ethernet segment? I'd recommend raising the mtu size on the
gigabit interfaces if this is the case. The default of 1500 is way too small
to hit peak performance between any too points simply because the machines
won't be able to send/receive that many packets per second. The sheer number
of interrupts needed to sustain that rate will kill just about any machine.
By increasing it to so called "jumbo frame" size (~9000 bytes) you can
increase your bandwidth tremendously. However, if that interface isn't
dedicated to speaking only to other jumbo frame configured devices, you'll
have a bit of a speed hit when the packets need to be fragmented (or worse,
you may encounter data not even being received in some circumstances).
-Eric Gross
"Vahid" <NOSPAM-sunman@ureach.com> wrote in message
news:uKWOb.21655$OM2.5360582@news4.srv.hcvlny.cv.net...
> Hi,
> We have tested transfer rate between two gige interface with only one
> gige switch in the middle. They are all connected via fiber cable.
> At the best performance, we get about 300mbits/s which is way lower than
> what it should be. Can anyone point me to some setting to check for? The
> following is what ever settings I can think of to include here:
> We are setting this at boot up:
> ndd -set /dev/ge instance 1
> ndd -set /dev/ge adv_1000autoneg_cap 1
> ndd -set /dev/ge adv_1000fdx_cap 1
> ndd -set /dev/ge adv_1000hdx_cap 0
>
> # ifconfig -a
> lo0: flags=849<UP,LOOPBACK,RUNNING,MULTICAST> mtu 8232
> inet 127.0.0.1 netmask ff000000
> ge1: flags=863<UP,BROADCAST,NOTRAILERS,RUNNING,MULTICAST> mtu 1500
> inet 192.168.73.101 netmask ffffff00 broadcast 192.168.73.255
> #
> # uname -a
> SunOS db01 5.6 Generic_105181-33 sun4u sparc SUNW,Ultra-Enterprise
> #
> ip_ignore_redirect 0
> tcp_conn_grace_period 500
> tcp_conn_req_max_q 128
> tcp_conn_req_max_q0 1024
> tcp_conn_req_min 1
> tcp_cwnd_max 262144
> tcp_fin_wait_2_flush_interval 675000
> tcp_ip_abort_cinterval 180000
> tcp_ip_abort_interval 480000
> tcp_keepalive_interval 7200000
> tcp_recv_hiwat 8192
> tcp_rexmit_interval_initial 3000
> tcp_rexmit_interval_max 60000
> tcp_rexmit_interval_min 400
> tcp_slow_start_initial 1
> tcp_time_wait_interval name is non-existent for this module
> for a list of valid names, use name '?'
> tcp_xmit_hiwat 8192
> #
> Thank you,
> Vahid.
>
- Next message: Darren Dunham: "Re: how to increase "NGROUPS_MAX (supplementary Groups)""
- Previous message: Tony Curtis: "Re: pkgadd problems"
- In reply to: Vahid: "low gigE transfer rate"
- Next in thread: Rick Jones: "Re: low gigE transfer rate"
- Reply: Rick Jones: "Re: low gigE transfer rate"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|