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

20030612: data feed problems (cont.)

>From: Unidata Support <address@hidden>
>Organization: UCAR/Unidata
>Keywords: 200306111400.h5BE0HLd016841 LDM-6 IDD firewall

Adam,

re:
>I forgot to mention this in the email i just wrote.  Our firewall
>policy is set that port 388 is open to all and has a guarantee limit of
>200 kilo bytes and a max limit of 400 kilo bytes.

OK.  If your firewall is really not slowing down LDM traffic, then the
problem is probably related to the Ethernet card in your machine, bad
cabling to your machine, a faulty router, or some other factor.
All of these would be hard to believe given the results from 'mtr'.

We did a timing test using scp to transfer the ldm-6.0.13.tar.Z
file from UCAR to tornado.  Here are the results:

  [ldm@tornado ~]$ scp address@hidden:/home/ftp/pub/ldm/ldm-6.0.13.tar.Z .
  The authenticity of host 'laraine.unidata.ucar.edu (128.117.140.62)' can't be 
established.
  RSA key fingerprint is 2e:66:5c:df:da:be:b5:87:db:99:c1:2f:37:da:71:b1.
  Are you sure you want to continue connecting (yes/no)? yes
  Warning: Permanently added 'laraine.unidata.ucar.edu,128.117.140.62' (RSA) to 
the list of known hosts.
  address@hidden's password: 
  ldm-6.0.13.tar.Z     100% |*****************************|   638 KB    00:29 

638 KB were transferred in 29 seconds.  This is a transfer rate of
22 KB per second, which is no where near the rate you have setup for
traffic to/from tornado.  So, connections to/from tornado are slow
regardless of whether the LDM is involved or not.

For comparison, I just did the exact same time timing test to an equivalent
machine at the University of Rio de Janeiro in Brazil, and the results
were:

  [ldm@brisa ldm]$  scp address@hidden:/home/ftp/pub/ldm/ldm-6.0.13.tar.Z .
  ldm-6.0.13.tar.Z     100% |*************************************|   638 KB    
00:04    

So, the transfer rate from UCAR to Brazil is 159.5 KB/s while to tornado
it is 22 KB/s.

>This was set many months ago to see if that was the reason for the
>latency which it did not help.

Your original requests to LSU combined a number of feeds into a single
request.  Given that it has proven to be impossible to get the entire
HDS feed contents to tornado (you are receiving about 20% of the data),
this would account for the latencies for the other data in that one
feed.  Now, the other feeds are getting to tornado with little to no
latency.

Given the tests above, we are convinced that there is something amiss
close to tornado.  Exactly what that is remains be seen.

Tom