regarding rcp/ftp processes

[remer]

hi,

i have few questions regarding rcp/ftp processes.

i have a processes that try to do a 3-4 ftp/rcp from a remote server which has a 3.9 mb transfer sec, but when the ftp is increasing upto 10 ftp/rcp processes my rate was drastically change to 1mb transfer rate per sec. my question are:

Does two many ftp/rcp processes affect my available NIC?

I have a 2 active nic (lan0 and lan4), how can i maximize the use of the two nic? or is there a way on how can i choose which nic my ftp/rcp processes will pass thru?

upon my checking lan0 has the most I/O usage.

thanks in advance.

[VK2COT · Accepted solution]

Hello,

a) Firstly, it seems you are running on
100Mbs LAN. If so, then throughput of
3.9 MBs is still relatively poor.

You should check both servers (remote and local).

For example, for HP-UX server, does the
following command report any errors:

netstat -i

b) verify that the LAN switches and
cards on the server side are set to
the same type of operation. For example,
100Mbs, duplex, autonegotiate)...

c) If you really want to increase network
throughput through a given IP address, the
HP-UX has Auto Port Aggregation (APA). APA
combines several physical NIC ports into
one logical IP connection.

It also allows a physical port, cable, or
switch port, to fail without losing the
connection. APA is not supported on all
Ethernet switches!

Finally, APA is a licensed product and
comes with a price tag but it is very
powerful and robust.

d) You did not specify if your two active
NICs reside on the same subnet (that would not be an officially approved solution if
APA is not used).

If those two NICs are on different subnets,
and you want to use them without any
fancy software, then simply set your
FTP and RCP commands to connect to
remote servers by pointing to different
IP addresses on the remote server.

Cheers,

VK2COT

[remer]

here are the config of the 2 nic:

Class I H/W Path Driver S/W State H/W Type Description
====================================================================
lan 0 0/0/0/1/0 igelan CLAIMED INTERFACE HP A7109-60001 PCI 1000Base-T Core
lan 4 1/0/4/1/0 igelan CLAIMED INTERFACE HP A6825-60101 PCI 1000Base-T Adapter


Interval: 1 NETWORK BY INTERFACE
Network In Packet Out Packet In KB Out KB
Idx Interface Type Rate Rate Rate Rate
--------------------------------------------------------------------------------
1 lan0 Lan 4602.0/4602.0 432.0/ 432.0 6780.0/6780.0 26.0/ 26.0
2 lan3 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
3 lan4 Lan 3.0/ 3.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
4 lan5 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
5 lan1 Lan 4.0/ 4.0 4.0/ 4.0 0.0/ 0.0 0.0/ 0.0
6 lan2 Lan 2.0/ 2.0 2.0/ 2.0 0.0/ 0.0 0.0/ 0.0
7 lo0 Loop na/ na na/ na na/ na na/ na



# ifconfig lan0
lan0: flags=1843
inet 10.131.3.101 netmask ffffff00 broadcast 10.131.3.255
# ifconfig lan4
lan4: flags=1843
inet 10.131.3.102 netmask ffffff00 broadcast 10.131.3.255

they reside on same subnet.

do you think i can use APA using those nic?

i tried to change the ip on the /etc/hosts and give the other nic (lan4) but it seems that lan0 has the highest util.


thanks.

[VK2COT]

Hello,

Uhh, you have both NICs on the same subnet.

HP-UX considers that two NICs configured
in the same subnet have equal connectivity.

Traffic will be received on either, but
will only be sent out through one or the other.

If you set ip_strong_es_model with ndd, it
may cause traffic that came in via one IP
have the replies go out that interface.
It will also require that traffic to a
given IP arrive on the interface with
the matching IP.

ip_strong_es_model:

Controls support for "Strong End-System
Model" described in RFC1122, Section 3.3.4.2.
When enabled, packet source addresses
(and therefore interfaces on a multihomed
host) affect selection of a gateway for
outbound packets. Set to 0 to disable;
set to 1 to enable. Default: 0 (disable)

Setting this value to one will have the
beneficial effect of allowing per-interface
default routes. It also means that if a
packet is received on a given interface,
the reply to that packet will be sent out
that interface. This can be useful if one
is in the rare situation of needing to have
separate physical (in the context of IP -
see ip_ill_status) interfaces configured
with IP addresses in the same subnet.

Generally, using Auto Port
Aggregation (APA) to create one virtual
interface with a logical interface for each address is more robust option.

Also, when ip_strong_es_model is set to a value of one, IP datagrams arriving on the
"wrong" interface (one that does not have an IP address which matches the IP datagrams'
destination IP address) are discarded. If
one is using IP address aliases on the loopback (lo0) interface in support of functionality such as hardware load balancer
triangle routing, setting ip_strong_es_model
to a value of one may result in loss of connectivity for the virtual IP address.

So, go for the APA if you have the budget to
purchase it.

Cheers,

VK2COT

[VK2COT]

Two more things:

a) What do the following command show:

lanadmin -x 0
lanadmin -x 4

b) Network statistics:

netstat -i

VK2COT

[remer]

here's the output.

# netstat -i
Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Coll
lan0 1500 10.131.3.0 10.131.3.101 1713862707 0 4244692507 0 0
lo0 4136 loopback localhost 40667406 0 40667406 0 0
lan4 1500 10.131.3.0 drppcsd1 3555452208 0 3702750702 0 0
# lanadmin -x 0
Speed = 100 Full-Duplex.
Autonegotiation = On.

# lanadmin -x 4
Speed = 100 Full-Duplex.
Autonegotiation = On.

#


btw, where can i find "ip_strong_es_model"?

thanks.

[VK2COT]

Hello again,

Your cards look good. There are no
collisions or errors on both interfaces.

Regarding ip_strong_es_model:

a) I am sure there are other posts in this Forum. Just search for them.

b) Here is a nice reference:

http://docs.hp.com/en/11890/perf-whitepaper-tcpip-v10.pdf

There are many more at docs.hp.com.

Cheers,

VK2COT

[Bill Hassell]

> ip_strong_es_model

This is one of the many tuning options available with the command ndd. To see all the options:

ndd -h

You can change values while running but you'll need to edit the nddconf file to make the change(s) permanent.

Concerning the transfer rate, are you measuring the total transfer rate for all the processes at the same time? 1 ftp process might run at 50-70 MB/sec but two will run at 25-35 MB each. Thus 10 processes will run about 5-7MB (probably slower due to the overhead in interleaving all the requests). Note that APA won't help if your other end is on a single 1GB network. For APA to improve total throughput, the pipes between the two machines must be big enough to keep all the APA cards busy. And of course, your local machine will run no faster than the remote server.