[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

20040216: LDM testing at SJSU

>From: Mike Voss <address@hidden>
>Organization: SJSU
>Keywords: 200402010010.i110ACp1023488 IDD test

Hi Mike,

>Our testing has progressed here at SJSU and some problems seem to have
>"mysteriously" dissappeared.

The number of times that problems have mysteriously disappeared
after demonstrating things like packet shaping to IT groups who
deny that anything is going on has to make you wonder :-)

>I'm now feeding methost24.met.sjsu.edu (my
>test machine) from emo off the same switch as I'm feeding
>rossby.met.sjsu.edu (my operational machine). Both of these machines
>have negotiated 100 full duplex connections, but methost24 is is
>keeping up with the entire feed while rossby can't handle it. Methost24
>is a P-III, 1 Ghz, 1 GB RAM machine while rossby is a SUN Ultra 60 with
>512 MB of RAM.

A Sun Ultra 60 should easily be able to keep up with the ingest
load that rossby has.  What we could do is instrument rossby to
see if some tuning would help.  We (Mike Schmidt) did several things
to tune the kernel on thelma.  But, then again, thelma is moving
a lot more data than rossby is receiving.

>Now, technically I'm not comparing apples to apples here because 1) the
>machines are different and 2) they are feeding from different machines
>at Unidata. If possible, for a short time I would like to switch the
>feeds so that rossby is feeding from emo and methost24 is feeding from
>thelma. This will allow me to eliminate thelma/emo differences as a
>cause of the differences I'm seeing between rossby and methost24.

I don't like to mess with the allow list on thelma as it requires that
we shutdown service, albeit temporarily, to 126-130 downstream
connections.  The good news is that I just checked the allows on thelma
and see that any machine at sjsu is allowed for all feeds except WSI
and NLDN.

>Is this reasonable from a practical standpoint and from a logic
>standpoint?

Yes.

>Or do you think I can gather from what I'm finding that
>rossby, my Sun Ultra 60, just can't handle the entire feed. It is doing
>no processing by the way, and I was under the assumption from
>conversations in the past that this machine should be able to handle
>everything.
>Thanks for your thoughts on this,

A Sun Ultra 60 should be able to ingest all of the data in the IDD
AND decode it.  It seems like there might be some tuning needed
on it, or perhaps there is some problem with the ethernet interface?

I will talk to Mike Schmidt about rossby today.

Tom
--