F11ACP Performance Measurements for RSX-11M
                                        
                               Joseph S. Sventek
                   Computer Science & Mathematics Department
                          Lawrence Berkeley Laboratory
                              Berkeley, CA  94720



        One  of  the  many  questions  presented  to  an  RSX-11M system
        generator during Phase II of SYSGEN is:
        
>* 11. FCP desired  (Type <ESC> for explanation)  [D: MIDDLE] [S]:<ESC>
>;
>;     Five FCP's are supported by SYSGEN.  They are:
>;
>;     Non-multiheader  - Minimum functionality (Size: 2K)
>;     Small            - Full functionality, heavily overlayed (2.25K)
>;     Middle           - Default, optimized for RMS, open/close envir. (5.2K)
>;     Large            - Optimized for all applications (6.7K)
>;     PLAS             - Fastest possible, optimized for create/delete (10.1K)
>;
        
        Little in the way of direct information  concerning  performance
        vs.   memory   occupancy  is  provided,  other  than  the  terse
        one-liners above. Thus, one is forced to rely upon the intuition
        that the larger ACP's should perform better, since less time  is
        spent  bringing  in  the  appropriate  overlay. As a result, the
        system generator usually opts for as large an ACP as the  system
        can tolerate in hopes of maximizing system performance.

        In  order  to  truly  "maximize system performance", a metric is
        needed to permit comparison of the performance of systems  which
        are identical except for the inclusion of different ACP options.
        Due  to  the  vastly  different  hardware  configurations  which
        populate the  RSX-11M  community,  it  is  difficult  to  devise
        metrics  which  give  accurate  absolute  measures; even so, one
        should be able to provide quantitative relative  measures  on  a
        particular  machine, and thereby aid the system generator in the
        choice of the option most appropriate to the particular system.

        The metric for ACP effectiveness used in  this  paper  is  based
        upon the following two principles:

          1. The  effectiveness of an ACP is large if the time necessary
             to perform a given benchmark is small. Symbolically,

                              effectiveness ~ 1 / time

          2. The ACP effectiveness is large  if  the  size  of  the  GEN
             partition is large. Symbolically,

                              effectiveness ~ size(GEN)

             where   size(GEN)   =  total  memory  -  (size(RSX_EXEC)  +
             size(ACP)).

                                      -1-                               


        Combining these two relationships, the effectiveness  metric  is
        defined as

                        effectiveness ~ size(GEN) / time

        In  the limit where the total memory of the system is large (eg.
        1 MB 11/70), the ACP effectiveness should be  dominated  by  the
        1/time term, since different sized ACP's will have little effect
        upon size of the GEN partition.

        It  is difficult to determine the constant of proportionality to
        arrive at an absolute effectiveness, but taking the ratio of one
        ACP's effectiveness to some standard ACP's effectiveness permits
        the determination of  the  relative  effectiveness.   Since  the
        MIDDLE   ACP   is  the  default  ACP  for  mapped  systems,  all
        effectivenesses listed in the tables and plotted in the  figures
        are relative to the effectiveness of a system with a single copy
        of the MIDDLE ACP.

        This  metric  has  been  used  to  compare several RSX-11M v.3.2
        systems with  different  ACP  options  included.   The  hardware
        configuration  for  the  tests  was an 11/34A CPU, 124KW of core
        memory, and an AED controller  driving  two  AMPEX  DM980  80-MB
        storage  modules.   The AED controller makes each storage module
        look like a 67MB RP03.  Two benchmark scripts were  run  on  the
        otherwise quiescent systems:

          1. DCHECK.CMD  -  Simultaneous wildcard PIP directory listings
             of the two disks are  performed.   This  simply  tests  the
             directory access portion of the ACP's.

          2. DWRITE.CMD  -  Several different PIP commands are performed
             simultaneously on the two disks, which is meant to exercise
             most of the overlays of the ACP's.

        Appendix A may be consulted for  the  exact  contents  of  these
        command files. In all cases, DP0: was the system disk, with 4490
        files  in  180 UIC's occupying 127000 blocks; DP1: was identical
        to DP0:.

        In order to calculate the effectiveness  of  the  ACP's,  it  is
        necessary  to calculate the size of the resulting GEN partition.
        For the measurements described  in  this  paper,  the  following
        partitions made up the RSX Executive:

                               EXEC        120000
                               LDRPAR        2500
                               DRVPAR        2000
                               TTPAR        20000
                               SYSPAR       10000

        This  corresponds  to  a  20K  executive, one disk driver, a 4KW
        full-duplex terminal driver, and space for MCR... and  the  task
        loader.   The  system  is  then  completed  by  inclusion of the
        particular ACP option, leaving the remainder of  memory  to  the
        GEN partition.

                                      -2-                               


        Four ACP's were tested in various configurations:

          1. MD1  - this is the ACP one gets when MIDDLE is specified to
             the SYSGEN question quoted  above.   This  is  usually  the
             default ACP for mapped systems.

          2. LG1  -  the  ACP one gets when the answer to question 11 is
             LARGE.

          3. RO2 - the ACP one gets when the answer to  question  11  is
             PLAS.

          4. CO1  -  this  is  a  core-only  ACP  which was derived from
             [1,24]fcpco0bld.cmd  and  [1,24]fcpcorbld.odl.   The  files
             [1,24]fcpco1bld.cmd and [1,24]fcpco1bld.odl are provided in
             appendix B.

        Tables  I  and  II  present  the  data  for  the  two benchmarks
        executing on systems with various ACP options. (Figures I and II
        are pictorial representations of the tables.)   The  first  four
        rows  of the tables present the data for systems which contain a
        single copy of the relevant ACP.  Two  relative  effectivenesses
        are  displayed:  one  for  a  124KW 11/34, and a second assuming
        512KW of memory, as in a  1MB  11/70.   The  total  memory  size
        affects the size of the GEN partition, and the "11/70" column is
        included to show that the relative effectiveness is dominated by
        the 1/time term.

        Two interesting trends can be observed from the data:

          1. For DWRITE, which is the more realistic of the two scripts,
             it is the case that the order of effectiveness is MD1 < LG1
             <  RO2.   On  an  18-bit  machine,  CO1  is  actually  more
             effective than RO2, while the opposite is true on a  22-bit
             machine.

          2. For  DCHECK,  the  resident overlaid ACP is actually slower
             than the disk overlaid ones, since it must dynamically  map
             itself to its data buffers.

        There  have long been rumors that RSX-11M systems perform better
        when each disk drive has a dedicated ACP.  When a single ACP  is
        servicing  multiple  disks, much of the directory context of one
        disk is lost if the ACP needs to service one of the other disks,
        with the result that when attention is returned to the  original
        disk,  that  context  must  be  re-established  via  extra  disk
        activity.  By dedicating an ACP/disk, this loss of  context  can
        be avoided.

        To test this hypothesis, four more systems were built, each with
        an   ACP   per   disk  drive,  using  MD1,  LG1,  RO2  and  CO1,
        respectively.  The second four rows of tables I and II show  the
        results  of the measurements on these systems.  The message from
        these measurements is clear:   it  is  much  more  effective  to
        dedicate  an  ACP  per disk, even on 18-bit machines.  On 18-bit
        machines, 2 copies of CO1 is more effective than RO2, while  the

                                      -3-                               


        reverse  is true for 22-bit machines.  In the case of DCHECK, it
        is still the case that the resident overlaid  system  is  slower
        than the others.

        Even  though  this metric indicates that it is more effective to
        use the extra memory for additional ACP's (due to the  increased
        speed of execution), owners of 18-bit machines may still balk at
        dedicating ~30000 bytes of memory for the file system, since the
        GEN  partition  can  become  intolerably small.  The only way in
        which multiple ACP's may appeal to these owners is if the memory
        cost/ACP becomes small.  This can  be  achieved  by  building  a
        multi-user F11ACP.

        A  multi-user  task  is one in which the pure code and pure data
        have  been  segregated  from  the  impure  code  and  data;  the
        resulting  read-only  segment  is  mapped by a separate APR.  In
        this way, many copies of the read-write portion of the task  can
        be  simultaneously  active,  mapped  over  a  single copy of the
        read-only segment.  If the size  of  the  read-only  segment  is
        represented  by  RO  and  that of the read-write is RW, then the
        memory savings by building the task multi-user is

                  n * (RW + RO) - (RO + n * RW) = (n - 1) * RO

        where  n  is  the  number  of  copies  of  the  task  that   are
        simultaneously active.

        An article by Eric Levy of JPL in the February 1981 Multi-tasker
        described the steps necessary to build a multi-user task for use
        under  RSX-11M.   The  most  recent distribution of the Software
        Tools Virtual Operating System (Spring 1981 RSX Symposium  Tape)
        provides  an  utility  (MUBLD)  which  performs  the  operations
        described in the article.  To  build  a  multi-user  version  of
        F11ACP, one must:

          1. Patch  TKB  as  per  the  June  1981  Software Dispatch for
             RSX-11M+ on pages 123-126.  These  patches  permit  TKB  to
             correctly  build  a  priveleged multi-user task.  Note that
             the correct checksums for the ".POB" modules are
                                 P2POP.POB     21461
                                 SGALO.POB     22734

          2. Assemble FCPMU.MAC provided in  appendix  C  below.   These
             PSECT  declarations  (which were provided by Larry Baker of
             USGS, Menlo Park, CA) serve to define the read-only psects.

          3. TKB @FCPMUBLD, using the patched version  of  TKB  and  the
             files  FCPMUBLD.CMD and FCPMUBLD.ODL provided in appendix C
             below.  This results in a task image  FCPMU.TSK  which  has
             the  RO  code  collected together and mapped by APR6.  This
             task image is input for MUBLD.  (The warning messages  from
             TKB  concerning  multiply  defined  PSECTS  can  be  safely
             ignored.)

          4. If it has not yet  been  done,  build  the  software  tools
             system  from  the  Spring  1981  RSX  SIG  Tape.  After the

                                      -4-                               


             TOOLGEN is completed, it is nessary to fix a bug  in  MUBLD
             with the following sequence of commands:

                  >"Get a copy of [307,33]mubld.w into the current directory"
                  >SHL
                  % ar xv mubld.w
                  % ar xv mubld.r domu
                  % ed domu
                  : /PIDSIZE/
                  : s//20/p
                  : w
                  : q
                  % ar uv mubld.r domu
                  % rc -v mubld.r
                  % logout
                  >

          5. >SHL -C MUBLD -V FCP
             Mubld  extracts  the RO  code into a resident-library style
             task image named rofcp.tsk.  It also generates another task
             image mufcp.tsk which has the RW code and is mapped over  a
             common partition named ROFCP for its RO code.

          6. Look  at  fcpmu.map  for  the  sizes of the two sections of
             code:
             
             R/W MEM LIMITS:  120000  135623  015624  07060.
             R-O MEM LIMITS:  140000  157277  017300  07872.

             (These are the numbers one should see if  the  EXTSCT's  in
             fcpmubld.cmd are as they appear in appendix C.  As such the
             size  of  the  RO  segment  is 017300 and that of the RW is
             015700.)

          7. Armed with the sizes of the two code sections,  VMR  a  new
             system  image with the appropriately sized partitions.  For
             example, if two ACP's are desired, with the device name  of
             the  second  disk  being  DB1:,  then  the  appropriate VMR
             commands are:

             SET /MAIN=ROFCP:*:173:COM
             INS ROFCP
             SET /MAIN=FCPPAR:*:336:SYS       ! 33600 = 2 * 15700
             INS MUFCP/TASK=F11ACP/CKP=NO
             INS MUFCP/TASK=DB01F1/CKP=NO

             The copies of mufcp must go into  a  system  partition,  as
             tasks  in  a  TASK partition cannot be mapped over a common
             region.

          8. After the initial boot of the system, execute

             >FIX F11ACP
             >FIX DB01F1

             before saving the system.  VMR does not  permit  fixing  of

                                      -5-                               


             tasks in SYS partitions.

          9. The  normal  mount of DB1: in startup.cmd will cause DB01F1
             to be used, since the order of ACP precedence when mounting
             a disk is:

               i. If /ACP=XXXXXX in MOU command, use XXXXXX
              ii. If this is ddnn: and DDNNF1 is installed, use DDNNF1
             iii. Use F11ACP


        A system with two copies of the multi-user  ACP  was  generated,
        and  the  same  two  benchmarks  were run.  The results of these
        benchmarks can be found in the ninth row of  tables  I  and  II.
        Since  MUFCP is simply a multi-user version of FCPCO1, it is not
        surprising that the  execution  times  are  identical  with  the
        system  with  two  copies  of CO1.  It is also apparent that the
        MUFCP system is the most effective ACP option  in  nearly  every
        situation.   The  author's standard system is now the multi-user
        system, and has been running for more than one month without any
        problems.

        In conclusion, it is true that in multiple disk systems (program
        development engines), an ACP/disk provides a  large  performance
        enhancement  over  a  single,  large ACP, whether disk or memory
        overlaid.  On memory-tight systems,  this  can  be  achieved  by
        building a multi-user version of F11ACP.






























                                      -6-                               


        










                                    Table I
                                        
                   ACP Effectiveness for DWRITE.CMD Benchmark
                                        
         System    Size(bytes)    Time(secs)    11/34 eff.    11/70 eff.
        ----------------------------------------------------------------
          MD1   |     10304    |     651     |    1.000    |    1.000
          LG1   |     13312    |     624     |    1.026    |    1.040
          RO2   |     20480    |     591     |    1.042    |    1.090
          CO1   |     14912    |     599     |    1.060    |    1.082
        2 * MD1 |     20608    |     401     |    1.534    |    1.606
        2 * LG1 |     26624    |     354     |    1.679    |    1.808
        2 * RO2 |     40960    |     318     |    1.713    |    1.982
        2 * CO1 |     29824    |     333     |    1.752    |    1.915
        2 * MU1 |     22080    |     334     |    1.827    |    1.925





                                    Table II
                                        
                   ACP Effectiveness for DCHECK.CMD Benchmark
                                        
         System    Size(bytes)    Time(secs)    11/34 eff.    11/70 eff.
        ----------------------------------------------------------------
          MD1   |     10304    |     971     |    1.000    |    1.000
          LG1   |     13312    |     971     |    0.984    |    0.997
          RO2   |     20480    |     991     |    0.927    |    0.969
          CO1   |     14912    |     971     |    0.975    |    0.995
        2 * MD1 |     20608    |     325     |    2.824    |    2.956
        2 * LG1 |     26624    |     324     |    2.737    |    2.946
        2 * RO2 |     40960    |     349     |    2.328    |    2.693
        2 * CO1 |     29824    |     326     |    2.669    |    2.918
        2 * MU1 |     22080    |     325     |    2.800    |    2.951











                                      -7-                               


                  Appendix A - Benchmark Command Files


DWRITE.CMD

INSTALL $PIP/TASK=...P0P
INSTALL $PIP/TASK=...P1P
TIME
.XQT P0P @DP0:[2,154]P0P
.XQT P1P @DP1:[2,154]P1P
.WAIT ...P0P
.WAIT ...P1P
TIME
REMOVE ...P0P
REMOVE ...P1P




DCHECK.CMD

INSTALL $PIP/TASK=...P0P
INSTALL $PIP/TASK=...P1P
TIME
.XQT P0P NL:=DP0:[*,*]/LI
.XQT P1P NL:=DP1:[*,*]/LI
.WAIT ...P0P
.WAIT ...P1P
TIME
REMOVE ...P0P
REMOVE ...P1P




DP0:[2,154]P0P.CMD

DP0:[2,377]=DP0:[307,31]*.*;*
NL:=DP0:[2,377]/FU
NL:=DP0:[2,377]*.CMD
DP0:[2,377]*.*/PU
DP0:[2,377]*.*;1=DP0:[2,377]*.*;*/RE
DP0:[2,377]*.*;*/DE




DP1:[2,154]P1P.CMD

DP1:[2,377]=DP1:[307,31]*.*;*
NL:=DP1:[2,377]/FU
NL:=DP1:[2,377]*.CMD
DP1:[2,377]*.*/PU
DP1:[2,377]*.*;1=DP1:[2,377]*.*;*/RE
DP1:[2,377]*.*;*/DE


                                      -8-                               


Appendix B - Files for Building FCPCO1


FCPCO1BLD.CMD

;       F C P C O 1 B L D . C M D
;
; BUILD CORE RESIDENT F11ACP (FCPCO1) FOR MAPPED RSX11M
;       NAME - FCPCO1
;       .ODL - FCPCORBLD.ODL
;       SIZE - 6. KW
;       FCB LIST - 32 FCB'S
;       SEPARATE DIRECTORY BUFFERS - 3
;
;
TK:[1,54]FCPCO1/AL/AC/MM/-FP,MP:[1,34]FCPCO1/-SP=FCPCO1BLD.ODL/MP
;
; OPTIONS INPUT
;
TASK=F11ACP
STACK=32
UNITS=1
UIC=[1,1]
PRI=149
PAR=FCPPAR:0:0
;       PAR=FCPPAR:0:30100
TSKV=.SSTVC:7
;
; THE MAXIMUM SIZE FOR THIS FCP IS 12K. ONLY THE DIRECTORY BUFFER, HOWEVER,
; CAN LIE IN THE LAST 4K (APR7).  THE FOLLOWING BUFFERS MUST BE IN THE LOWER 8K
; (APR5 AND APR6).
;
;
; ALLOCATE A SEPARATE BUFFER FOR INDEX FILE BITMAP
;
;       EXTSCT=$$BUF0:1006
;
; DEFAULT IS ONE BUFFER SHARED BETWEEN BITMAP AND FILE HEADERS (AND DIRECTORY
; BLOCKS).  THIS ACTION WILL SPEED UP APPLICATIONS THAT DO A LOT OF FILE
; CREATION AND DELETION.
;
;
;
; ALLOCATE A SEPARATE BUFFER FOR FILE HEADERS
;
EXTSCT=$$BUF1:1006
;
; DEFAULT FROM ASSEMBLY IS ONE BUFFER SHARED BETWEEN FILE HEADERS
; AND DIRECTORY BLOCKS (AND INDEX FILE BITMAP).  THIS ACTION WILL BENEFIT ANY
; APPLICATION THAT CAN AFFORD THE MEMORY.
;
;
;
; ALLOCATE PRIVATE FCB POOL SPACE
;
; EACH FCB = 52 OCTAL

                                      -9-                               


;
;       EXTSCT=$$AFR1:XXXX
;
;       ,WHERE XXXX = N*52 (OCTAL) AND N = NUMBER FCB'S DESIRED
;
EXTSCT=$$AFR1:2500
;
;
; *** NOTE ***
; THE PRECEDING BUFFERS AND EXTSCT'S MUST LIE IN ADDRESS SPACE BELOW APR7.
; DO NOT EXTSCT PAST 157777 OCTAL!!
;
; THE DIRECTORY BUFFER BELOW MAY LIE, IN WHOLE OR IN PART, WITHIN THE ADDRESS
; SPACE OF APR7.  CONSEQUENTLY, THE MAXIMUM ALLOWABLE FOR THIS EXTSCT IS
; <3.5K + N> OR <17000 (OCTAL) + N> WHERE N REPRESENTS THE AMOUNT OF SPACE UNDER
; APR7 THAT IS STILL AVAILABLE AFTER THE PRECEDING EXTSCT'S HAVE BEEN APPLIED.
;
;
; ALLOCATE ADDITIONAL BUFFER SPACE FOR DIRECTORY BLOCKS
;
;       EXTSCT=$$BUF3:XXXX
;
;       ,WHERE XXXX REPRESENTS 1000 FOR EACH 'ADDITIONAL' BLOCK
;       DESIRED.
;
; DEFAULT FROM ASSEMBLY IS ONE DIRECTORY BLOCK.  NOTE THAT THIS
; ACTION SHOULD BE IN ADDITION TO ALLOCATING A SEPARATE BUFFER
; FOR FILE HEADERS.  ALLOCATING ADDITIONAL BUFFER SPACE FOR DIRECTORY
; BLOCKS WITHOUT A SEPARATE BUFFER FOR FILE HEADERS DOES NOT
; ACCOMPLISH VERY MUCH.
;
EXTSCT=$$BUF3:2000
;
;
;
/





FCPCO1BLD.ODL

;       F C P C O R B L D . O D L
;
;       OVERLAY DESCRIPTION FOR CORE RESIDENT F11ACP - RSX11M/M+
;
;               MINIMUM SIZE -- 6.0/6.25 K  PHYSICAL
;               MAXIMUM SIZE -- 12 K  PHYSICAL
;
;       THIS ODL BUILDS A CORE RESIDENT F11ACP.  IT DOES NOT REQUIRE PLAS.
;       IT IS INTENDED FOR USE IN SYSTEMS WITHOUT PLAS OR WITH APPLICATIONS
;       THAT WANT A SMALL RESIDENT F11ACP (THIS ODL DOES NOT REQUIRE
;       A SEPARATE BUFFER FOR FILE HEADERS LIKE RESIDENT OVERLAY ODL'S).
;       NOTE THAT IT PROVIDES A MINIMUM OF SPACE AVAILABLE FOR FCB'S.
;

                                      -10-                              


;
;       STRUCTURE:
;               APR5 - ROOT - CODE
;               APR6 - ROOT - CODE
;               APR6 - CO-TREE - BUFFERS
;               APR7 - CO-TREE - DIRECTORY BUFFER  (MAY OVERMAP I/O PAGE)
;
;
        .NAME   F11ACP
        .ROOT   F11ACP-FCPROT-FCPCO
;
;       THE FOLLOWING MODULES MUST BE IN THE ROOT
FCPROT: .FCTR   FCP/LB:F11ACP:F11CM:SMCOM:DISPAT-FCPRT2
FCPRT2: .FCTR   FCP/LB:ARWVB:OVERR:FCPGBL-CODE
;
;       CO-TREE CONTAINING DIRECTORY BUFFER OVERMAPS APR7
FCPCO:  .FCTR   FCP/LB:F11BUF:DIRBUF:INIT
;
;       THESE ARE THE REST OF THE MODULES FOR FCP
CODE:   .FCTR   FCP/LB:ALLOC:BLXIO:CKSUM:CLACC:CLDAC:CLDIR:CRFID-CODXX1
CODXX1: .FCTR   FCP/LB:DARITH:DLMRK:DRACC:DREOF:DREX:DRGET:DRINI-CODXX2
CODXX2: .FCTR   FCP/LB:DRPAC:DRSEF:DRVLB:DRWRT:DWPND:FDRMV:FNDNM-CODXX3
CODXX3: .FCTR   FCP/LB:GTFID:GTMAP:INFCB:INWIN:LOCAT:MPHDR:MPVBN-CODXX4
CODXX4: .FCTR   FCP/LB:NXHDR:PROCK:RDHDR:RLEAS:RLFCB:RWVB:RWVBL:RW1LB-CODXX5
CODXX5: .FCTR   FCP/LB:SCFAC:SCFCB:SMALC:SMNXB:SMRVB:SMSCN:WACCK:WITRN-CODXX6
CODXX6: .FCTR   FCP/LB:WRHDR:WTRN1:ACCESS:ATCTL:CRFIL:DATIM:DEACC:ENTNM-CODXX7
CODXX7: .FCTR   FCP/LB:EXCMP:EXCOM:EXTEN:RATCM:RDATT:RWATT:WATCM:WRATT-CODXX8
CODXX8: .FCTR   FCP/LB:CLDEL:CLATT:CLCRE:CLEXT:CLFCB:CLNUP:DLBLK:DLFIL-CODXX9
CODXX9: .FCTR   FCP/LB:DLHDR:DRALC:DRCPY:DREXT:EXTHD:IXEXT:RMVNM:SMDEL-CODX10
CODX10: .FCTR   FCP/LB:TRUNC:DMOUNT-[1,1]EXELIB/LB-[1,54]RSX11M.STB/SS
;
        .END

























                                      -11-                              


Appendix C - Files for Building Multi-user F11ACP


FCPMU.MAC

        .title  fcpmu
        .ident  /1.0/

        .psect  access  ro,i,lcl,rel,con
        .psect  alloc   ro,i,lcl,rel,con
        .psect  atctl   ro,i,lcl,rel,con
        .psect  blxio   ro,i,lcl,rel,con
        .psect  cksum   ro,i,lcl,rel,con
        .psect  clatt   ro,i,lcl,rel,con
        .psect  clcre   ro,i,lcl,rel,con
        .psect  cldac   ro,i,lcl,rel,con
        .psect  cldel   ro,i,lcl,rel,con
        .psect  cldir   ro,i,lcl,rel,con
        .psect  clext   ro,i,lcl,rel,con
        .psect  clfcb   ro,i,lcl,rel,con
        .psect  clnup   ro,i,lcl,rel,con
        .psect  crfid   ro,i,lcl,rel,con
        .psect  crfil   ro,i,lcl,rel,con
        .psect  datim   ro,i,lcl,rel,con
        .psect  deacc   rw,i,lcl,rel,con
        .psect  dispat  rw,i,lcl,rel,con
        .psect  dlblk   ro,i,lcl,rel,con
        .psect  dlfil   ro,i,lcl,rel,con
        .psect  dlhdr   ro,i,lcl,rel,con
        .psect  dlmrk   ro,i,lcl,rel,con
        .psect  dmount  ro,i,lcl,rel,con
        .psect  dracc   ro,i,lcl,rel,con
        .psect  dralc   ro,i,lcl,rel,con
        .psect  drcpy   rw,i,lcl,rel,con
        .psect  dreof   ro,i,lcl,rel,con
        .psect  drex    ro,i,lcl,rel,con
        .psect  drext   rw,i,lcl,rel,con
        .psect  drget   ro,i,lcl,rel,con
        .psect  drini   ro,i,lcl,rel,con
        .psect  drpac   ro,i,lcl,rel,con
        .psect  drsef   ro,i,lcl,rel,con
        .psect  drvlb   ro,i,lcl,rel,con
        .psect  drwrt   ro,i,lcl,rel,con
        .psect  dwpnd   ro,i,lcl,rel,con
        .psect  entnm   ro,i,lcl,rel,con
        .psect  excmp   ro,i,lcl,rel,con
        .psect  excom   ro,i,lcl,rel,con
        .psect  exten   ro,i,lcl,rel,con
        .psect  exthd   ro,i,lcl,rel,con
        .psect  fcpgbl  ro,i,lcl,rel,con
        .psect  fdrmv   ro,i,lcl,rel,con
        .psect  fndnm   ro,i,lcl,rel,con
        .psect  f11acp  ro,i,lcl,rel,con
        .psect  f11cm   rw,i,lcl,rel,con
        .psect  gtfid   ro,i,lcl,rel,con
        .psect  gtmap   ro,i,lcl,rel,con

                                      -12-                              


        .psect  infcb   ro,i,lcl,rel,con
        .psect  inwin   rw,i,lcl,rel,con
        .psect  ixext   ro,i,lcl,rel,con
        .psect  locat   rw,i,lcl,rel,con
        .psect  mphdr   ro,i,lcl,rel,con
        .psect  mpvbn   ro,i,lcl,rel,con
        .psect  nxhdr   ro,i,lcl,rel,con
        .psect  prock   ro,i,lcl,rel,con
        .psect  ratcm   ro,i,lcl,rel,con
        .psect  rdatt   ro,i,lcl,rel,con
        .psect  rdhdr   ro,i,lcl,rel,con
        .psect  rleas   ro,i,lcl,rel,con
        .psect  rlfcb   ro,i,lcl,rel,con
        .psect  rmvnm   ro,i,lcl,rel,con
        .psect  rwatt   ro,i,lcl,rel,con
        .psect  rwvb    ro,i,lcl,rel,con
        .psect  rwvbl   rw,i,lcl,rel,con
        .psect  rw1lb   rw,i,lcl,rel,con
        .psect  scfac   ro,i,lcl,rel,con
        .psect  scfcb   ro,i,lcl,rel,con
        .psect  smalc   ro,i,lcl,rel,con
        .psect  smcom   rw,i,lcl,rel,con
        .psect  smdel   ro,i,lcl,rel,con
        .psect  smnxb   ro,i,lcl,rel,con
        .psect  smrvb   ro,i,lcl,rel,con
        .psect  smscn   ro,i,lcl,rel,con
        .psect  trunc   ro,i,lcl,rel,con
        .psect  wacck   ro,i,lcl,rel,con
        .psect  watcm   ro,i,lcl,rel,con
        .psect  witrn   ro,i,lcl,rel,con
        .psect  wratt   ro,i,lcl,rel,con
        .psect  wrhdr   ro,i,lcl,rel,con
        .psect  wtrn1   rw,i,lcl,rel,con
        .psect  dirbuf  ro,i,lcl,rel,con
        .psect  f11buf  ro,i,lcl,rel,con
        .psect  init    ro,i,lcl,rel,con
        .psect  $$afr1  rw,i,lcl,rel,con
        .psect  $$afr2  rw,i,lcl,rel,con
        .psect  $$buf0  rw,i,lcl,rel,con
        .psect  $$buf1  rw,i,lcl,rel,con
        .psect  $$buf2  rw,i,lcl,rel,con
        .psect  $$buf3  rw,i,lcl,rel,con
        .psect  $$buf4  rw,i,lcl,rel,con
        .psect  $$buf5  rw,i,lcl,rel,con

        .end





FCPMUBLD.CMD

;
;       f c p m u b l d . c m d
;

                                      -13-                              


; build multi-user f11acp task (fcpmu) for rsx-11m
; this task image can then be munged with mubld to generate the images
; necessary for installation in an rsx-11m system
;       name - fcpmu
;       .odl - fcpmubld.odl
;       size - ??? kw
;       fcb list - 32. fcb's
;       separate directory buffer - 3 blocks
;
;
fcpmu/al/ac/mm/-fp/mu,fcpmu/-sp=
@fcpmubld.odl
;
; options input
;
task=f11acp
stack=32
units=1
uic=[1,1]
pri=149
par=fcppar:0:0
tskv=.sstvc:7
;
; the maximum size for this fcp is 12k. only the directory buffer, however,
; can lie in the last 4k (apr7).  the following buffers must be in the lower
; 8k (apr5 and apr6)
;
;
; allocate a separate buffer for index file bitmap
;
;       extsct=$$buf0:1006
;
; default is one buffer shared between bitmap and file headers (and directory
; blocks).  this action will speed up applications that do a lot of file
; creation and deletion.
;
;
;
; allocate a separate buffer for file headers
;
extsct=$$buf1:1006
;
; default from assembly is one buffer shared between fil headers
; and directory blocks (and index file bitmap).  this action will benefit any
; application that can afford the memory
;
;
;
; allocate private fcb pool space
;
; each fcb = 52 octal
;
;       extsct=$$afr1:xxxx
;
;       ,where xxxx = n * 52 (octal) and n = number of fcb's desired
;

                                      -14-                              


extsct=$$afr1:2500
;
;
; *** note ***
; the preceding buffers and extsct's must lie in address space below apr7.
; do not extsct past 157777 octal!!
;
; the directory buffer below may lie, in whole or in part, within the
; address space of apr7. consequently, the maximum allowable for this extsct
; is <3.5k + n> or <17000 (octal) + n> where n represents the amount of space
; under apr7 that is still available after the preceding extsct's have been
; applied.
;
;
; allocate additional buffer space for directory blocks
;
;       extsct=$$buf3:xxxx
;
;       ,where xxxx represents 1000 for each 'additional' block desired.
;
; default from assembly is one directory block.  note that this
; action should be in addition to allocating a separate buffer for file
; headers.  allocating additional buffer space for directory blocks
; without a separate buffer for file headers does not accomplish very much.
;
extsct=$$buf3:2000
;
;
;
/



FCPMUBLD.ODL

;       f c p m u b l d . o d l
;
;       note:   this is not a real .odl file, but is referenced in the build
;               files in a line as @fcpmubld.odl
;
fcpmu
[1,24]fcp/lb:f11acp:f11cm:smcom:dispat:arwvb:overr:fcpgbl
[1,24]fcp/lb:f11buf:dirbuf:init
[1,24]fcp/lb:alloc:blxio:cksum:clacc:cldac:cldir:crfid
[1,24]fcp/lb:darith:dlmrk:dracc:dreof:drex:drget:drini
[1,24]fcp/lb:drpac:drsef:drvlb:drwrt:dwpnd:fdrmv:fndnm
[1,24]fcp/lb:gtfid:gtmap:infcb:inwin:locat:mphdr:mpvbn
[1,24]fcp/lb:nxhdr:prock:rdhdr:rleas:rlfcb:rwvb:rwvbl:rw1lb
[1,24]fcp/lb:scfac:scfcb:smalc:smnxb:smrvb:smscn:wacck:witrn
[1,24]fcp/lb:wrhdr:wtrn1:access:atctl:crfil:datim:deacc:entnm
[1,24]fcp/lb:excmp:excom:exten:ratcm:rdatt:rwatt:watcm:wratt
[1,24]fcp/lb:cldel:clatt:clcre:clext:clfcb:clnup:dlblk:dlfil
[1,24]fcp/lb:dlhdr:dralc:drcpy:drext:exthd:ixext:rmvnm:smdel
[1,24]fcp/lb:trunc:dmount
[1,1]exelib/lb,[1,54]rsx11m.stb/ss
/

                                      -15-