CDE and slow redraw times

[Peter A. Berger Jr.]

I know HP doesn't support Hummingbird Exceed 8.0 (but someone out there has to be running it) running on a new HP-UX 11.11 (RP-7410, 3GB RAM, 733Mhz CPU's).
The screen redraws are horribly slow on this primary server, but not on an identically setup development server.
Using the same Exceed program on both servers. Is there anything I can optimize? I don't see much. local video = 1024x768. Is there perhaps a CDE setting within HP-UX I can check too?
Thanks in advance. :)

[Jeff Schussele]

Hi Peter,

Check name resolution of your display system from each of these servers. Note any return time discrepancies between them.
Then check overall network performance from these servers to your system. Try to ftp BIG files between the server & you or better yet use ethereal to evaluate the network perf.
Sounds like a possible name resolution or network performance issue to me.

Rgds,
Jeff

[Robert Gamble]

Are the network settings OK on the primary server OK ? port negotiation ok ? I've seen poor network performance cause the symptoms you are describing ...

[Jim Mallett]

I had a redraw problem with GUI applications on a new system before, not CDE but with others.

I made sure that my settings in the following were correct, (the new machine still had the default settings):
/etc/resolv.conf
/etc/nsswitch.conf

Once corrected, the GUIs came up like a charm.

Jim

[Tim Sanko]

One thing they haven't mentioned was the switch settings if they are managed and you have a mismatch, it could really pile up errors and retries.

Check with lanadmin the card statistics. Check the switch port for errors, and check for speed/duplexing mismatch.

Tim

[Peter A. Berger Jr.]

RP-7140 plugged directly into core switch (HP 4108gl) and all ports reporting zero errors and all at auto-negeotiation at 100mbs.
/etc/hosts checked and resolved find.
/etc/resolv.conf and /etc/nsswitch set correctly.
CPU and NIC utlization not high at all. Memory hovers around 98-100% util (this is production whereas the machine that handles GUI apps fine doesn't have nearly as many people on it). Could this high % of memory util. be the key for GUI application (especially Data Protector 5.0) window redraw slowdowns? Thanks.

[Jeff Schussele]

HI Peter,

Check the actual port duplex setting.
lanadmin -x 0 #if lan0 use 1 for lan1, etc.
If you're running half-duplex you'll have collisions naturally as well as using only half the pipe. If you're full-duplex, then you're OK. I *never* trust auto-negotiation - I lock port & switch at 100 & full-duplex.

Now - you're memory. Just how is that memory being allocated? If it's mainly User & Buffer, that's probably OK - you're just using it. Might as well - you paid for it. If there's a significant portion allocated to System AND you have high page-out rates (which should reflect in high swap usage) then you're exhausting memory & you need to determine why.
Could be you need more or could be you need to look at kernel parms (dbs_max_pct?) to release more memory for applications.

HTH,
Jeff

[Peter A. Berger Jr.]

Since it's a production machine I'll wait until downtime to force 100FDx.
As for memory see attachment (sysdef062503.txt). dbc_max_pct = 15 and dbc_min_pct = 5.
Most of the memory util. is indeed U & B.

[Peter A. Berger Jr.]

Since it's a production machine I'll wait until downtime to force 100FDx.
As for memory see attachment (sysdef062503.txt). dbc_max_pct = 15 and dbc_min_pct = 5.
Most of the memory util. is indeed U & B.

[Jeff Schussele]

Hi (again) Peter,

Yes that would be wise to wait for maint.
Forcing the switch to 100FD will cause about a 10 second network outage which is normally OK UNLESS you have something sensitive on that NIC (MC/SG heartbeat?)
But DON'T forget to have that switch locked into 100FD when you make that switch so it won't try to negotiate 100HD again.
And to memory, you should really start looking thoroughly at just how memory is being used.
Does vmstat 5 10 output show *any* values for po (page out)? How many? If it's showing double digits you're probably memory bound.
Also check for memory leaks. Do processes seem to *grow* over time slowly? If yes they're bleeding memory & need to be fixed.
Search the forum for past posts on determining memory usage & you'll find tons of execellent threads.

Rgds,
Jeff