36f8f6fb024b20cc523cdf9ebde7491eca84fd7c
markd
  Sun Dec 6 20:33:20 2020 -0800
multiple request per connect works except hgBlat

diff --git src/gfServer/README.blat src/gfServer/README.blat
index 7076ac2..86e881d 100644
--- src/gfServer/README.blat
+++ src/gfServer/README.blat
@@ -1,70 +1,86 @@
 This archive should contain everything you need
 to make a BLAT server.
 
 There are seven directories:
    lib - Library routines.  
    inc - Interface to library routines
    hg/lib - Human genome project specific library
    hg/inc - Human genome project specific interfaces
    gfServer - Source to 'server' that sits in about
               a gigabyte of memory listening to a port
 	      for queries.  GF = Genomic Finder.
    gfClient - Source to a client that operates in text
               mode from the command line.
    hg/hgBlat - Source to a client that operates on the
                Web and links to UCSC browser.
 
 To get things going read how to get the code to compile
 in the general purpose README, following the general
 install and then the BLAT install instructions.
 Then ftp over to genome-ftp.soe.ucsc.edu, go to
 goldenPath/07oct2000/bigZips, and download chromNib.tar.Z.
 This contains the Oct 7th assembly in a format that
 puts two bases per byte, and is randomly accessible.
 Log onto the machine that you want to become the server.
 Create a directory where you want this to go, go
 into this directory, and gtar -Zxf chromNib.tar.Z.
 Then do:
     gfServer
 to review the gfServer command line options and
     gfServer start hostName 17777 *.nib &
 The gfServer will take about 10 minutes to index
 the files (first counting 12-mers, then actually
 indexing).  For 'hostName' you want to fill in the host
 name of the server on your side.  For us it's 
 kks00.soe.ucsc.edu.  You don't have to use the
 port 17777 - that's the one we use though.
 
 After the server says it's ready for queries
 you can test it a little.  A quick test is:
     gfServer files hostName 17777 
 which should just print a list of all the .nib files
 and the DNA sizes.  You can do further testing
 with an .fa file, including the ones that
 are part of the gfServer directory.
 
 The gfClient will produce a file in the '.psl'
 format we use for alignments.  I think for
 'industrial' use I'd want to change the client
 and the server so that they didn't have to
 set up a new connection for each query.
 I really built this for the web where this
 overhead doesn't matter, but it would add
 up if you were trying to process a million
 reads or ESTs with gfClient.  Not that
 you *do* need to have the nib files in a
 directory that is accessible from the client
 as well as the server.
 
 hgBlat is currently hard wired to query
 kks00.soe.ucsc.edu.  Please change this
 to your local server once things are up
 if you are mirroring the web site.  
 
 Thanks much for your interest, and best
 of luck with it.  Let me know how it 
 goes.
 	-Jim Kent  March 12, 2001
 
 
 
+Dynamic BLAT server:
+
+gfServer now supports a dynamic mode, which allows servers that are
+started on demand without needing to continually occupy memory.  A single
+dynamic gfServer can serve multiple genomes and multiple simultaneous
+users.
+
+An index is prebuilt using 'gfServer index'.  See gfServer help
+for details and naming convention.  Fast hard disk or SSDs should be
+used to store the genomes and indexes.  Sufficient memory will be
+need for unique genome indexes that are used simultaneously.  Queries
+against we same genome index will be shared between multiple process.
+
+The server is then configued to start via xinetd.  See
+blat-xinetd.example for an example configuration.
+