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

20030212: needed CONDUIT feed pattern modifications

>From: Unidata Support <address@hidden>
>Organization: UCAR/Unidata
>Keywords: 200212271733.gBRHXAt25394 LDM 6 regular expression

Hi Art,

Profiling of the LDM has revealed a major bottleneck in the regular
expression code on Sun Solaris (both SPARC and x86).  In particular,
Steve Emmerson determined that a Sun will work a lot harder (something
like a factor of 1300) with split feed requests that look like:

request NMC2    ".*[09]$"   nmc2_0
request NMC2    ".*[18]$"   nmc2_1
request NMC2    ".*[27]$"   nmc2_2
request NMC2    ".*[36]$"   nmc2_3
request NMC2    ".*[45]$"   nmc2_4

than the following feed requests:

request NMC2    "[09]$"   nmc2_0
request NMC2    "[18]$"   nmc2_1
request NMC2    "[27]$"   nmc2_2
request NMC2    "[36]$"   nmc2_3
request NMC2    "[45]$"   nmc2_4

Steve is investigating H. Spencer's regular expression package to see
if using it on Sun gets rid of/mitigates this bottleneck.  Until
then, we are requesting that you change your CONDUIT feed requests on
ldm.meteo to the second set listed above and then stop and restart your
LDM (assuming that the aliases you are using for thelma.ucar.edu, which
_is_ a Sun Enterprise 450 running Solaris) are nmc2_0, etc.).

Just to let you know, the investigation that led to the finding above
is part of the development of LDM-6.  We have deployed test versions of
this new LDM on all machines under our direct control, and several
others as well -- Harry Edmon of U Wash. is using LDM 6 on the UW
machine that is requesting CONDUIT data from thelma --.  We havn't asked
you to switch to this new version yet since we havn't done enough
testing of it on Linux.  Those tests will be coming up soon.

Thanks in advance...

Tom

>From address@hidden Wed Feb 12 15:39:04 2003

Tom,

I've made the changes as requested below.

                        Art.