Slow network transfer speed Tru64<>HP-UX

[Bojan SER]

Hello everyone,

i'm having a problem with ftp copy speeds between Tru64 (5.1) and HP-UX (11.31)...
I have two 8640 and one gs1280, boxes are not under load, APA/LAG have been set on all 3 of them (dual 1gbit cards on every box)... when I try to copy from Alpha to either of integrity servers I get 500-800kbyte/s ftp transfer
but when I try to copy between 2 integrity servers I get speeds around 120 mbyte/s)...

did anyone had similar problem?

[Steven E. Protter]

Shalom,

Can not help much on the Tru64 side.

But.

lanadmin -x #
Replace # with the number from nwmgr or lanscan output.

You are looking for 1000 BaseT Auto negotiate environment.

Based on your notes, I'd say the area to concentrate is on the Tru64 side.

See if mii-tool or ethtool are supported in the environment to report network speeds.

SEP

[Steven Schweda]

> [...] Tru64 (5.1) [...]

Actual output from "sizer -v" would provide
more details.

To get the current interface characteristics:

hwmgr get attribute -category network

To set something, see:

man lan_config

> [...] ftp copy speeds [...]

Is FTP the only slow thing? Have you tried,
say, NFS? SCP?

I've never done any link aggregation, so I
know nothing, but you might see if using a
single (simple) interface is slower (or not).

[rick jones]

I would not expect APA/LAG to increase the performance of a single stream beyond the capacity of a single link in the aggregate. Unless the systems are back-to-back it will be the switch that determines the packet scheduling on inbound.

One of my favorite suspects in a poor network performance situation is packet loss, so I would suggest checking the statistics on either side of the transfer.

After that, I look at things like TCP window sizes.

[Bojan SER]

hello all, thanks for replies...
>sizer -v
HP Tru64 UNIX V5.1B (Rev. 2650); Fri May 7 09:26:58 CEST 2010

tried scp also, same speed 500-800kbyte/s...
also tried with non aggregated links, same result...
copying between two integrity servers works smooth and fast so I think it's not them...
network statistics output is in attached file...

[Steven Schweda · Accepted solution]

Is there any more network hardware between a
Tru64 system and an HP-UX system than there
is between two HP-UX systems, or is
everything plugged into one big switch?

How fast is a Tru64 system when it's talking
to itself? (Is the disk (or something else
on the Tru64 system) slow, or is the network
between systems slow?)

[Elmar P. Kolkman]

It looks like your network links on the Tru64 system is running at 100Mb/s. Or that some other network part is limiting the speed to that 10Mb/s.

First make sure that the link between the HP-UX boxes is the same as between the Tru64 and the HP-UX's. Then make sure that the Tru64 server is running its link at 1Gb and not something slower. (I don't know commands for that, on HP-UX it's lanadmin).

[Steven Schweda]

> It looks like your network links on the
> Tru64 system is running at 100Mb/s.

Which "it" looks like that? This?:

> media_speed = 1000

Or this?:

> media_speed = 1000

Or something else?


> [...] I don't know commands for that, [...]

Read more?

[rick jones]

Netstat (or netstat like - aka TCP) statistics would be nice - or did I not correctly ID that in the attachment (I'm an HP-UX/Linux type not Tru64)?

Is the MTU 9000 bytes on all systems and the switches between all systems?

[Bojan SER]

Problem solved... it was a emc storage connected to Alpha with poor performance... tried to copy from ram disk and went fine...
then connected alpha to new flash array and wohoo... 120-140mbyte/s alpha<>integrity over dual gbit aggregated link...

thanks for replies
this can be closed now