AIX 5.2 ML01 - slow GIGABIT write

[MoshiachNow]

HI,

I have a 6E3 machine with an onboard GIGABIT adapter.
My disk is 72 GB IBM.
The average "dd" throughput to this disk is around 50 MB/sec.
When I run a rcp of a 1GB file to a 6F1 (also AIX 5.2) machine - it shows abot 40MB/sec (same if I rcp this file from 6F1 to 6E3).
When I pull this file back - I get only 9 MB/sec.
Clearly it's not a disk issue - hence the good "dd" performance.

Playing with mbuf and some "no" parameters does not seem to help.

Anybody has got a similar experience?
I'm starting to suspect some ML issues (?).
================================
no -a|grep -E "tcp|thew"
tcp_bad_port_limit = 0
tcp_ecn = 0
tcp_ephemeral_high = 65535
tcp_ephemeral_low = 32768
tcp_finwait2 = 1200
tcp_init_window = 0
tcp_inpcb_hashtab_siz = 24499
tcp_keepcnt = 8
tcp_keepidle = 14400
tcp_keepinit = 150
tcp_keepintvl = 150
tcp_limited_transmit = 1
tcp_maxburst = 0
tcp_mssdflt = 512
tcp_nagle_limit = 65535
tcp_ndebug = 100
tcp_newreno = 1
tcp_nodelayack = 0
tcp_pmtu_discover = 1
tcp_recvspace = 245760
tcp_sendspace = 245760
tcp_timewait = 1
tcp_ttl = 60
thewall = 1048576
==========================




"Long live king Moshiach !"

http://www.7for70.com/

 

[MoshiachNow]

P.S.

Jumbo frames set on ent+switch.

"Long live king Moshiach !"

http://www.7for70.com/

 

[MoshiachNow]

Have found the reason - 6E3 and 6E4 come with both system and data disks connected to the same scsi channel ...
Running the same test to the externally connected tower gave around 50 MB/sec !
This explains why 6F1 with the same version and network interface type was so much faster - the system and data disks are splited there on different scsi busses.

"Long live king Moshiach !"

http://www.7for70.com/