Return to share.5 CVS log

Up to [Research Unix] / researchv10dc / man / manx

Annotation of researchv10dc/man/manx/share.5, revision 1.1

1.1     ! root        1: .TH SHARE 5 SHARE
        !             2: .SH NAME
        !             3: Share \- Share Scheduling on Unix
        !             4: .SH SYNOPSIS
        !             5: Scheduling for a share of the machine
        !             6: .SH DESCRIPTION
        !             7: .I Share
        !             8: is a term covering those elements of the Unix kernel
        !             9: that affect the priority of a user's job.
        !            10: The basic scheduler in Unix schedules processes on a short term
        !            11: .I per-process
        !            12: basis.
        !            13: The
        !            14: .I "share scheduler"
        !            15: takes account of the history of a user's
        !            16: usage of the resources of the machine, and introduces a
        !            17: .I per-user
        !            18: long term scheduler.
        !            19: To do this, several variables are available to the scheduler in a per-user data
        !            20: structure known as an
        !            21: .I lnode.
        !            22: These record the intended share of the machine that the user should get,
        !            23: the recent history of resources consumed (``usage''),
        !            24: and the number of active (running) processes belonging to the user.
        !            25: Together, these affect the priority of each of the processes
        !            26: so that consumption of resources is adjusted toward the intended share.
        !            27: .P
        !            28: A user's
        !            29: .I usage
        !            30: is calculated by accumulating the charges incurred by use of resources,
        !            31: and decaying the result over time.
        !            32: The share scheduler affects the low-level scheduling of a user's processes
        !            33: by adding the user's
        !            34: .I usage
        !            35: divided by the allocated share,
        !            36: and multiplied by the active process count,
        !            37: to the priority of a process
        !            38: every time that process incurs a clock tick.
        !            39: Since the larger its priority,
        !            40: the less often a process is scheduled,
        !            41: processes belonging to users with high
        !            42: .IR usage ,
        !            43: low share, or many active processes
        !            44: will get a smaller share of resources.
        !            45: Note that at any one time, a user can use all of the resources available
        !            46: provided there is no competition from others.
        !            47: .SS "SCHEDULING GROUPS"
        !            48: .I lnodes
        !            49: are organised into a tree.
        !            50: For any particular sub-tree,
        !            51: the sub-tree's share of resources
        !            52: is divided up between the
        !            53: lnodes
        !            54: according to their relative shares.
        !            55: Sub-trees are also known as groups.
        !            56: The root
        !            57: .I lnode
        !            58: of the group is the group owner,
        !            59: and the leaf
        !            60: .I lnodes
        !            61: are its users.
        !            62: The total shares issued to the group include both those issued to the owner,
        !            63: as well as those issued to the users.
        !            64: Both owner and users are known as group members.
        !            65: The share of the group's resources
        !            66: allocated to any particular member of the group
        !            67: is in the proportion
        !            68: of the member's shares divided by the group's shares.
        !            69: .P
        !            70: The most interesting group is the one at the top of the tree,
        !            71: whose owner is ``root'',
        !            72: and its group members are the primary scheduling groups.
        !            73: \&``root''
        !            74: gets 100% of the available resources,
        !            75: which is split as above between the primary scheduling groups.
        !            76: .P
        !            77: Not all group owners represent real users,
        !            78: and in these cases there is no need to allocate them a share of resources.
        !            79: Such lnodes are indicated by the \s-1NOTSHARED\s0 flag,
        !            80: which causes the scheduler to ignore their shares when allocating
        !            81: their group's resources among its members.
        !            82: However,
        !            83: the long term
        !            84: .I charge
        !            85: of a group owner
        !            86: always includes all the charges levied on any member of its group.
        !            87: .P
        !            88: For reasons of system management,
        !            89: \&``root'' is always allocated 100% of the resources whenever it needs them.
        !            90: However, since all real users run on their own lnodes,
        !            91: the \s-1NOTSHARED\s0 flag is turned on for ``root'',
        !            92: and thus the primary scheduling groups
        !            93: have 100% of the available resources to share between themselves.
        !            94: However, for instrumentation purposes,
        !            95: ``root''s
        !            96: .I charge
        !            97: only represents its own consumption of resources,
        !            98: but the total consumption of resources is accumulated in the
        !            99: .I kl_temp
        !           100: field of
        !           101: \&``root''s
        !           102: .I kern_lnode
        !           103: (see
        !           104: .IR lnode (5)
        !           105: for details of a
        !           106: .IR kern_lnode .)
        !           107: .SS CHARGES
        !           108: Charges making up the accumulating usage figure
        !           109: are levied by default as follows:-
        !           110: .RS
        !           111: .PD 0
        !           112: .TP "\w'system servicesXX'u"
        !           113: cpu
        !           114: 100%
        !           115: .TP
        !           116: disk i/o
        !           117: 0%
        !           118: .TP
        !           119: terminal i/o
        !           120: 0%
        !           121: .TP
        !           122: system services
        !           123: 0%
        !           124: .TP
        !           125: memory
        !           126: 0%
        !           127: .PD
        !           128: .RE
        !           129: .P
        !           130: Memory charges are levied every scheduler cycle,
        !           131: but note that 
        !           132: .I root
        !           133: is never charged for the memory it uses.
        !           134: These charges can be varied at different times of the day to reflect
        !           135: their popularity by using the command
        !           136: .IR charge (1).
        !           137: .P
        !           138: Usage is decayed at an exponential rate intended to ensure that
        !           139: all users of the machine get an equal chance to compete for resources
        !           140: over a particular time period.
        !           141: The default decay results in a
        !           142: .I half-life
        !           143: of 2 minutes.
        !           144: Use
        !           145: .IR charges (1)
        !           146: to find out the current decay rate and resource charges.
        !           147: .SS NICE
        !           148: The
        !           149: .IR nice (2)
        !           150: system call has a slightly different effect under Share.
        !           151: The
        !           152: .I nice
        !           153: parameter for a process now affects the rate at which its priority decays
        !           154: to a higher priority over time.
        !           155: .I Nicing
        !           156: a process will make it run slower,
        !           157: by reducing its effective share of the resources,
        !           158: but it may not run slower than another user's processes
        !           159: if that user has an even lower effective share of the resources.
        !           160: However, processes with a
        !           161: .I nice
        !           162: priority of 19 are guaranteed only to run when no other processes need the CPU.
        !           163: .I Niced
        !           164: processes are charged less for CPU time than normal processes,
        !           165: priority 19 processes are charged almost nothing for CPU time.
        !           166: .SS MANAGEMENT
        !           167: There are three flags that control the operation of the share scheduler.
        !           168: .TP "\w'LIMSHAREXX'u"
        !           169: .SM NOSHARE
        !           170: This turns off the scheduler.
        !           171: Since this will leave the parameters in a ``frozen'' state,
        !           172: it should probably only be done at system boot time.
        !           173: This flag is on by default when the system is booted,
        !           174: so the command
        !           175: .IR charge (1)
        !           176: must be run to activate the scheduler.
        !           177: Note that the program
        !           178: .IR login (8)
        !           179: won't attach users to their own
        !           180: .I lnodes
        !           181: while this flag is on,
        !           182: instead each user will remain attached to
        !           183: .IR root 's
        !           184: lnode.
        !           185: [Value 01]
        !           186: .TP
        !           187: .SM ADJGROUPS
        !           188: This flags turns on global group effective share adjustment.
        !           189: If any group is found to be getting less than
        !           190: .SM MINGSHARE
        !           191: times its allocated share,
        !           192: then the costs incurred by its members are reduced proportionately.
        !           193: [Value 02]
        !           194: .TP
        !           195: .SM LIMSHARE
        !           196: This flag deals with an 
        !           197: ``edge effect''
        !           198: that occurs when users first log in.
        !           199: It may be that their
        !           200: .I usage
        !           201: field has decayed to the point where they might temporarily be allocated
        !           202: nearly 100% of the machine.
        !           203: This flag limits any one user's share of the resources to no more than
        !           204: MAXUSHARE times their intended share.
        !           205: Of course, this still may be nearly 100%, if no other users are logged in,
        !           206: or the other users have very small shares.
        !           207: [Value 04]
        !           208: .P
        !           209: The
        !           210: .IR charge (1)
        !           211: command is used to manipulate these flags, and the charging parameters above.
        !           212: There are also other parameters which may be changed with
        !           213: .IR charge :-
        !           214: .TP "\w'MAXNORMUSAGEXX'u"
        !           215: .BI DecayRate
        !           216: The decay rate for users'
        !           217: .IR "active process rates" .
        !           218: This parameter is calculated by counting the active proceses
        !           219: per user every clock scan,
        !           220: and is decayed every clock scan.
        !           221: The usual value for this should result in a
        !           222: .I half-life
        !           223: for the rate of about 10 seconds.
        !           224: .TP
        !           225: .BI DecayUsage
        !           226: The decay rate for users' usages.
        !           227: This may be altered to produce a
        !           228: .I half-life
        !           229: for usage ranging from a few seconds to many days.
        !           230: .TP
        !           231: .BI Delta
        !           232: The run frequency of the share scheduler in seconds.
        !           233: The default value of 4 is fine.
        !           234: .TP
        !           235: .BI MAXGROUPS
        !           236: This sets the maximum group nesting (depth of scheduling tree) allowed,
        !           237: not including ``root''s group.
        !           238: .TP
        !           239: .BI MAXPRI
        !           240: Absolute upper bound for a process's priority.
        !           241: .TP
        !           242: .BI MAXUPRI
        !           243: Upper bound for normal processes' priorities.
        !           244: .I Idle
        !           245: processes run with priorities in the range
        !           246: \s-1MAXUPRI\s0 < \fIpri\fP < \s-1MAXPRI\s0.
        !           247: .TP
        !           248: .BI MAXUSAGE
        !           249: Upper bound for ``reasonable'' usages.
        !           250: Users with usages larger than this are grouped together and given
        !           251: process priorities
        !           252: which prevent them from interfering with ``normal'' users.
        !           253: The usage
        !           254: (multiplied by the
        !           255: .IR "active process rate" )
        !           256: is added to a running process's priority
        !           257: every time it incurs a clock tick,
        !           258: so the upper bound should be small enough not to overrun the value
        !           259: .SM MAXUPRI
        !           260: in too short a time interval
        !           261: .TP
        !           262: .BI MAXUSERS
        !           263: Sets the maximum number of users and groups that can be active.
        !           264: Note that this cannot exceed the maximum configured in the kernel.
        !           265: .TP
        !           266: .BI PriDecay
        !           267: This is the decay rate for maximally niced processes.
        !           268: A reasonable minimum value for the
        !           269: .I half-life
        !           270: is about 100 seconds,
        !           271: but see the comment for 
        !           272: .SM MAXUSAGE
        !           273: above.
        !           274: .TP
        !           275: .BI PriDecayBase
        !           276: The base for calculating the decay rate
        !           277: for process priorities with normal \fInice\fP.
        !           278: This should be set low enough so that the priorities of processes
        !           279: for users with low share don't decay too quickly.
        !           280: A reasonable minimum value for the
        !           281: .I half-life
        !           282: is about 2 seconds.
        !           283: .SH FILES
        !           284: .PD 0
        !           285: .TP "\w'/usr/include/sys/charges.hXX'u"
        !           286: .IR /usr/include/sys/share.h
        !           287: Definition of scheduler parameters.
        !           288: .TP
        !           289: .IR /usr/include/sys/charges.h
        !           290: Default scheduler parameters.
        !           291: .PD
        !           292: .SH "SEE ALSO"
        !           293: charge(1),
        !           294: pl(1),
        !           295: rates(1),
        !           296: shstats(1),
        !           297: ustats(1),
        !           298: lnode(5),
        !           299: shares(5),
        !           300: login(8),
        !           301: sharer(8).
        !           302: .SH REFERENCES
        !           303: "Scheduling for a Share of the Machine", J Larmouth, SP&E, Vol 5 1975 pp 29-49
        !           304: .br
        !           305: "Scheduling for Immediate Turnaround", J Larmouth, SP&E, Vol 8 1978 pp 559-578
        !           306: .br
        !           307: "A Fair Share Scheduler", J Kay & P Lauder, TM 11275-870319-01
        !           308: .br
        !           309: "Share Scheduler Administration", P Lauder

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.