The size of the memory required for each unit of a HiRDB parallel server configuration is the sum of the items listed in the following table.

For details about the value specified in the shmmax operating system parameter (for Solaris, shmsys:shminfo_shmmax; for Linux, SHMMAX), see Chapter 20. Specifying OS Parameters.

If you increase the size of shared memory, it might affect transaction performance by increasing the number of page faults. For each operand, consult the guidelines for your particular system and then specify the most appropriate value.

Table 15-7 Size of memory required for each unit of a HiRDB parallel server configuration

Item				Required memory (KB)
Process private area	Process private area used by all unit controller processes			32-bit mode J + K number of FESs in unit + L (number of BESs in unit + number of DSs in unit) + {(64 + 48 (v + 1)) (value of pd_max_server_process - w - 3) + (64 + 48 (ac + 1)) 3 + z} 1,024 64-bit mode J + K number of FESs in unit + L (number of BESs in unit + number of DSs in unit) + {(64 + 64 (v + 1)) (value of pd_max_server_process - w - 3) + (64 + 64 (ac + 1)) 3 + z} 1,024 + ab If a plug-in is used, add: + 1,400 If the asynchronous READ facility is used, add: + s If Real Time SAN Replication is used, add: + (A + B + C) 1,024 If fixed is specified in the pd_process_terminator operand, add: + M (value of pd_process_terminator_max - 1) If you are performing in-memory data processing, add: + {T (value of pd_max_bes_process 2 + 7) number of BESs in unit} 1,024 If you are changing the maximum number of communication traces stored, add: + V 1,024
	Process private area used by each server process#1, #2	Front-end server		(N + h + m + p + q) b + y If you are changing the maximum number of communication traces stored, add: + W 1,024
		Dictionary server	pd_work_buff_mode=each specified	{P + i + m + (a + 9) 2 + r + t} b + y + S If you are changing the maximum number of communication traces stored, add: + W 1,024
			pd_work_buff_ mode=pool specified or omitted	32-bit mode (P + i + m + a + a 128 0.1 + 11 + n + r + t) b + y + S 64-bit mode (P + i + m + a + a 128 0.1 + 15 + n + r + t) b + y + S If you are changing the maximum number of communication traces stored, add: + W 1,024
		Back-end server	pd_work_buff_ mode=each specified	{Q + g + (a + 9) c + i + m + r + t} (b + 3) + y + S If you are performing in-memory data processing, add: + {T (b + 3)} 1,024 If you are changing the maximum number of communication traces stored, add: + W 1,024
			pd_work_buff_ mode=pool specified or omitted	32-bit mode (Q + g + a + a 128 0.1 + 11 + i + m + n + r + t) (b 3) + y + S 64-bit mode (Q + g + a + a 128 0.1 + 15 + i + m + n + r + t) (b 3) + y + S If you are performing in-memory data processing, add: + {T (b + 3)} 1,024 If you are changing the maximum number of communication traces stored, add: + W 1,024
Shared memory	Space used by the unit controller in the unit controller shared memory			d 1,024
	Space used by each server in the unit controller shared memory#1			e
	Global buffer shared memory			F
	In-memory data processing shared memory			U
	Utility shared memory			u
	Security audit information buffer shared memory			For automatic calculation by the system: 0.3 + MAX{(R + 100), (R 1.2)} 0.25 For user-specified values (specify the pd_audit_def_buffer_size operand): Value specified for pd_audit_def_buffer_size
	Inter-process memory communication shared memory#3			j k

#1: If the unit contains multiple servers (excluding the system manager), obtain the value for each server.

#2: When using plug-ins, add 300 per server process.

#3: Add this value if you have specified PDIPC=MEMORY in the client environment definition. For details about the inter-process memory communication facility and client environment definitions, see the HiRDB Version 9 UAP Development Guide. If either the HiRDB server or the HiRDB client is in 32-bit mode, the system allocates the shared memory for the inter-process memory communication facility in the 32-bit address space.

a: Value of pd_work_buff_size operand

b: Value of pd_max_users operand + value of pd_max_reflect_process_count operand

• For a dictionary server, the value is (value of the pd_max_dic_process operand + value of the pd_max_reflect_process_count operand).
• For a back-end server, the value is (value of the pd_max_bes_process operand + value of the pd_max_reflect_process_count operand).
• If the pd_max_dic_process or pd_max_bes_process operand is omitted, the value is (value of the pd_max_users operand + value of the pd_max_reflect_process_count operand).

c: Maximum number of work tables

Find the number of work tables for each SQL statement in Table 15-8 Procedure for obtaining the number of work tables for each SQL statement. Use the largest number of work tables obtained from Table 15-8 Procedure for obtaining the number of work tables for each SQL statement as the maximum number of work tables.

d: Value obtained from 15.2.3 Formulas for shared memory used by a unit controller.

e: Value obtained from 15.2.4 Formulas for shared memory used by each server.

f: Value obtained from 15.2.5 Formula for size of shared memory used by global buffers.

g: Size of memory required during SQL execution

For details about the formula, see 15.2.6 Formulas for size of memory required during SQL execution.

h: Size of memory required during SQL preprocessing

For details about the formula, see 15.2.7 Formula for size of memory required during SQL preprocessing.

i: LOB buffer batch input/output work memory

Add 62 KB if LOB global buffer is specified for the LOB RDAREA for the corresponding server (-b specified in the pdbuffer operand of the system common definition).

j: Maximum number of concurrently executable clients that use the inter-process memory communication facility.

If you are not sure about the value, specify the number of all clients that use the inter-process memory communication facility or the value of the pd_max_users operand.

k: Average memory size for data transfer performed by all clients that use the inter-process memory communication facility (value of PDSENDMEMSIZE + value of PDRECVMEMSIZE in the client environment definition).

m: Memory requirement for a Java virtual machine

If you use Java stored procedures or Java stored functions, add the size of memory used by the Java virtual machine. This value depends on the Java virtual machine's options (-Xms, -Xmx, and -Xmn options for Hewlett-Packard JRE 1.2.2.04) and version. For details about the memory requirement for your Java virtual machine, see the applicable manual. Following are the guidelines for the memory required for HP-UX:

• Eight MB of memory is required to start a Java virtual machine.
• Add the maximum memory size for the Java virtual machine (value of the -Xmx option). Note that some Java virtual machines may use more memory than the size specified in the -Xmx option.

n: Work table extended memory size

When the pd_work_buff_expand_limit operand is specified, add the work table extended memory size. The work table extended memory size is determined from the following formula:

Work table extended memory size (kilobytes) = work table extended buffer size + (work table extended buffer size 128) 0.1

• Work table extended buffer size (kilobytes) = MAX(0, work table extended buffer size based on hash join, subquery hash execution) + MAX(0, work table extended buffer size based on the increase in the number of work tables)
• Work table extended buffer size based on hash join, subquery hash execution = MIN{ (work table extended buffer size based on hash join, subquery hash execution - value of the pd_work_buff_size operand), (value of the pd_work_buff_expand_limit operand - value of the pd_work_buff_size operand) } number of concurrently executing users executing hash join, subquery hash execution
  For details about determining the work table extended buffer size when executing hash joins, subquery hash executions, see the HiRDB Version 9 UAP Development Guide.
  
• Work table extended buffer size based on the increase in the number of work tables = MIN{ (number of work tables used 128 - value of the pd_work_buff_size operand), (value of pd_work_buff_expand_limit operand - value of pd_work_buff_size operand) } (number of users such that the number of work tables is greater than the value of the pd_work_buff_size operand 128)
  Number of work tables used = MAX(number of work table files used per SQL statement, number of work table files used by the ASSIGN LIST statement)
  For details about determining the number of work table files used per SQL statement and the number of work table files used by the ASSIGN LIST statement, see 18.3 Determining the maximum number of files (pdfmkfs -l command).
  

p: Memory requirements required for BLOB data type

For details about the formula, see 15.2.8 Formula for size of memory required during BLOB data retrieval or updating (front-end server).

q: Memory requirements required for server-side block transfer or array FETCH

For details about the formula, see 15.2.10 Formula for size of memory required during block transfer or array FETCH (front-end server).

r: Memory requirements required for BLOB data type

For details about the formula, see 15.2.9 Formula for size of memory required during BLOB data retrieval or updating (back-end server or dictionary server).

s: Memory size used by asynchronous READ

This is applicable when the asynchronous READ facility is used; use the following formula (in kilobytes) for the calculation:

(90 + 90 Memory used by the RDAREA for management of the HiRDB file system area) i=1 value of pd_max_ard_process

For the memory used by the RDAREA for management of the HiRDB file system area, use the largest 90 areas in descending order of the values. If the number of areas used by the server is fewer than 90, assume that amount anyway.

The memory used by the RDAREA for management of the HiRDB file system area (in kilobytes) is calculated from the formula below based on the initial settings:

Note that the parameters at the time of initialization of the area can be checked by running the pdfstatfs command with the -A option specified.

{(Number of files#1 + number of extensions#2)64} 1.5#3

#1: Value specified by pdfmkfs -l, or displayed in available file count in the execution results of the pdfstatfs command.

#2: Value specified by pdfmkfs -e, or displayed in available expand count in the execution results of the pdfstatfs command.

#3: Multiply when the area size (value specified in pdfmkfs -n) is at least 2,048.

t: HiRDB file system memory size

Determine with the following formula (in kilobytes):

347 + Memory used by the work tables for management of the HiRDB file system area + Memory used by the system logs for management of the HiRDB file system area + 90 memory used by the RDAREA for management of the HiRDB file system area i=1

The memory used by the HiRDB file system area for management of work tables and system logs uses the maximum value calculated for the memory used by the HiRDB file system area for management used by the server. For RDAREAs, use the largest 90 areas in descending order of the values. If the number of areas used by the server is fewer than 90, use as many areas as are used for the calculation.

The memory used by the RDAREA for management of the HiRDB file system area (in kilobytes) is calculated with the formula below based on the initial settings:

Note that the parameters at the time of initialization of the area can be checked by running the pdfstatfs command with the -A option specified.

{(Number of files#1 + number of extensions#2)64} 1.5#3

#1: Value specified by pdfmkfs -l, or displayed in available file count in the execution results of the pdfstatfs command.

#2: Value specified by pdfmkfs -e, or displayed in available expand count in the execution results of the pdfstatfs command.

#3: Multiply when the area size (value specified in pdfmkfs -n) is at least 2,048.

u: When value of pd_utl_exec_mode is 0: 0

When value of pd_utl_exec_mode is 1: {(b 2,000 + 136) 1,024} 1,024

v: Value of pd_module_trace_max that is valid as the unit control information definition

w: Sum of (maximum number of processes that can be started + 3) for all server processes in the unit

For details about the maximum number of processes that can be started, see the manual HiRDB Version 9 System Definition.

y: Sum of the values obtained by the following formula for each server process in the unit:

In the 32-bit mode:

{(64 + 48 (value of pd_module_trace_max + 1)) (maximum number of processes that can be started + 3)} 1,024

In the 64-bit mode:

{(64 + 64 (value of pd_module_trace_max + 1)) (maximum number of processes that can be started + 3)} 1,024

For details about the maximum number of processes that can be started, see the manual HiRDB Version 9 System Definition.

z: Memory size for restarting HiRDB

If this memory size cannot be allocated, HiRDB restart fails. Use the following formula to determine the size (in bytes):

(D + E + F) number of dictionary servers + (D + E + F + G) number of back-end servers + H

Use the following variables for the formula to calculate the size of memory used by HiRDB to restart:

Variable	Value
D	32-bit mode 246,762 + 4 value of pd_max_rdarea_no + {48 (value of pd_max_rdarea_no + number of tables) + 304} ((value of pd_max_users# + value of pd_max_reflect_process_count) 2 + 7) 64-bit mode 305,274 + 8 value of pd_max_rdarea_no + {64 (value of pd_max_rdarea_no + number of tables) + 512} ((value of pd_max_users# + value of pd_max_reflect_process_count) 2 + 7) Number of tables: 62 + MAX{value of pd_max_access_tables, 500}
E	b1 X + b2 Y b1: When the record length of the server status file < 4,096 MAX(((3,400 ((((record length - 40) - 308) 20) + ((record length - 40) 20) (MAX(4,096 record length,2) - 1)) + 0.7)),1) MAX(4,096 record length,2) (record length - 40) When 4,096 record length of server status file < 12,288 MAX((3,400 ((((record length - 40) - 308) 20)) + 0.7),1) (record length - 40) When 12,288 record length of server status file MAX((3,400 ((((record length - 40) - 836) 20)) + 0.7),1) (record length - 40) X: When the number of RDAREAS in server 3,400: 1 When 3,401 number of RDAREAS in server 6,800: 2 When 6,801 number of RDAREAS in server: 3 b2: When the record length of the status file for server < 4,096 ((5,662,310 ((((record length - 40) - 308) 20) + ((record length - 40) 20) (MAX(4,096 record length,2) - 1)) + 0.7)) MAX(4,096 record length,2) (record length - 40) When 4,096 record length of server status file < 12,288 (5,662,310 ((((record length - 40) - 308) 20)) + 0.7) (record length - 40) When 12,288 record length of server status file (5,662,310 ((((record length - 40) - 836) 20)) + 0.7) (record length - 40) Y: When the number of RDAREAS in server 10,200: 0 When 10,201 number of RDAREAS in server 5,672,510: 1 When 5,672,511 number of RDAREAS in server 11,334,820: 2 When 11,334,821 number of RDAREAS in server: 3
F	If commit is specified in the pd_dbsync_point operand, add: + 112 ((value of pd_max_users# + value of pd_max_reflect_process_count) 2 + 7)
G	If 1 or a greater value is specified in the pd_inner_replica_control operand, add: 32-bit mode (48 value of pd_max_rdarea_no + 80) ((value of pd_max_users# + value of pd_max_reflect_process_count) 2 + 7) 64-bit mode (64 value of pd_max_rdarea_no + 160) ((value of pd_max_users# + value of pd_max_reflect_process_count) 2 + 7)
H	For back-end servers in which the number of HiRDB file system areas that store RDAREAs created in a character special file in a server is 1,001 or more, add: 32-bit mode 12,012 ((number of HiRDB file system areas that store RDAREAs created in a character special file - 1,000) 1,000) 64-bit mode 16,016 ((number of HiRDB file system areas that store RDAREAs created in a character special file - 1,000) 1,000)

#

For a dictionary server, use the value of pd_max_dic_process. For a back-end server, use the value of pd_max_bes_process. However, if both pd_max_dic_process and pd_max_bes_process are omitted, use the value of pd_max_users

A: 425 (2 b + 7) Number of FESs in unit

B: 425 (2 b + 7) Number of DSs in unit

C: 425 (2 b + 7) Number of BESs in unit

J, K, L, M, N, P, Q: Fixed value

These values depend on the OS being used. The following table presents the values for each OS (in kilobytes):

OS	Value of J	Value of K	Value of L	Value of M	Value of N	Value of P	Value of Q
HP-UX (32-bit mode)	80,000	9,700	12,600	2,800	4,700	3,800	3,600
HP-UX (32-bit-mode POSIX library version)	127,300	10,300	12,900	3,000	5,200	4,300	5,000
HP-UX (64-bit mode)	80,700	10,300	13,400	2,800	5,100	4,300	4,000
HP-UX (IPF)	200,500	15,100	17,900	2,600	7,100	8,100	10,600
AIX (32-bit mode)	67,700	8,400	10,700	2,200	5,100	4,100	4,300
AIX (32-bit-mode POSIX library version)	123,700	16,800	19,200	5,000	7,900	7,300	7,400
AIX (64-bit mode)	123,700	25,000	31,200	6,000	11,900	11,700	13,400
Solaris (32-bit mode)	71,100	6,100	7,600	1,400	2,600	2,100	4,100
Solaris (32-bit-mode POSIX library version)	121,100	6,800	8,200	1,700	2,800	2,300	4,400
Solaris (64-bit mode)	106,900	9,700	11,900	12,100	4,100	3,300	6,400
Linux (32-bit mode)	74,200	6,700	8,800	1,700	2,900	4,300	4,300
Linux (EM64T)	134,600	24,100	31,700	7,600	13,300	15,400	14,700
Linux 6 (64-bit x86_64)	106,400	12,300	14,600	1,900	5,400	5,800	7,200

R: If you are making a rough estimate, this is the number of audit events (the number of executions of CREATE AUDIT). If you are making a detailed estimate, it is the number of entries in the security audit information buffer

S: Memory required when using the facility for acquiring syncpoint output synchronization control information (bytes)

If 1 is specified in the pd_dbbuff_trace_level operand and the pd_dfw_awt_process operand is not specified, add:

32-bit mode

320 number of global buffers defined in a single server

64-bit mode

640 number of global buffers defined in a single server

T: If 1 or a greater value is specified in the pd_max_resident_rdarea_no operand, add:

1,648 + 16 value of pd_max_resident_rdarea_no + 16 value of pd_max_resident_rdarea_shm_no

U: Memory required by in-memory data processing

For the applicable formulas, see 15.2.11 Memory required by in-memory data processing.

V: Memory required by communication trace processing

32-bit mode

(16 (Z - 1,024) 2) (value of pd_max_server_process - w)

64-bit mode

(32 (Z - 1,024) 2) (value of pd_max_server_process - w)

W: Memory required by communication trace processing

This is either of the following values calculated for each server process within the unit.

32-bit mode

(16 (aa - 1,024) 2) (maximum number of startup processes + 3)

64-bit mode

(32 (aa - 1,024) 2) (maximum number of startup processes + 3)

For details about the maximum number of startup processes, see the manual HiRDB Version 9 System Definition.

Z: The value of pd_pth_trace_max enabled as the unit control information definition.

The value specified for the operand rounded up to a power of two.

aa: The value of pd_pth_trace_max enabled as each server definition.

The value specified for the operand rounded up to a power of two.

ab: Size of memory for signal handler

HP-UX (IPF) version: 768

Otherwise: 0

ac: If the operand pd_module_trace_max is specified in the system common definition or the unit control information definition: value of pd_module_trace_max

Otherwise: 16,383

Table 15-8 Procedure for obtaining the number of work tables for each SQL statement

SQL statement	Procedure for obtaining the number of work tables
SELECT statement INSERT(-SELECT) statement	When none of 1-8 as follows are applicable: 0 When any of 1-8 as follows are applicable: Sum of the applicable values from 1-8 When multiple tables are joined for retrieval Number of additional work tables = (Number of joined tables - 1) 2 + 1 When specifying the ORDER BY clause Number of additional work tables = 2 When specifying the GROUP BY clause Number of additional work tables = Number of GROUP BY clauses specified When specifying the DISTINCT clause Number of additional work tables = Number of DISTINCT clauses specified When specifying the UNION, UNION ALL, or EXCEPT[ALL] clause Number of additional work tables = (Number of UNION or UNION ALL clauses specified) 2 + 1 When search condition contains columns with index defined Number of additional work tables = Number of columns with index defined in the search condition When specifying the FOR UPDATE or FOR READ ONLY clause Number of additional work tables = 1 When specifying a subquery (quantified predicate) Number of additional work tables = Number of subqueries specified
UPDATE statement DELETE statement	Number of columns with index defined in the search condition + 1
DROP SCHEMA statement DROP TABLE statement DROP INDEX statement CREATE INDEX statement REVOKE statement to revoke access privilege	2