Return to session-management.tex CVS log

Up to [local] / OpenXM / doc / calc2000

Annotation of OpenXM/doc/calc2000/session-management.tex, Revision 1.1.1.1

1.1       noro        1: % $OpenXM: OpenXM/doc/issac2000/session-management.tex,v 1.13 2000/01/17 08:50:57 noro Exp $
                      2:
                      3: \section{Session Management}
                      4: \label{secsession}
                      5: %MEMO: key words:
                      6: %Security (ssh PAM), initial negotiation of byte order,
                      7: %mathcap, interruption, debugging window, etc.
                      8:
                      9: In this section we explain our control integration in
                     10: OpenXM.  We assume that various clients and servers
                     11: establish connections dynamically and communicate to each
                     12: other. Therefore it is necessary to give a dynamical and unified
                     13: method to start servers and to establish connections.
                     14: In addition to that, interruption of execution and
                     15: debugging facilities
                     16: are necessary for interactive distributed computation.
                     17:
                     18: %\subsection{Interface of servers}
                     19: %
                     20: %A server has additional I/O streams for exchanging data between
                     21: %a client and itself other than ones for diagnostic
                     22: %messages. As the streams are for binary data,
                     23: %the byte order conversion is necessary when a
                     24: %client and a server have different byte orders. It is determined by
                     25: %exchanging the preferable byte order of each peer. If the preference
                     26: %does not coincide with each other,
                     27: %then the network byte order is used.
                     28: %This implies that all servers and clients should be able to
                     29: %handle the network byte
                     30: %order. Nevertheless it is necessary to negotiate the byte order to
                     31: %skip the byte order conversion because its cost is often dominant over
                     32: %fast networks.
                     33:
                     34: \subsection{Invocation of servers}
                     35: \label{launcher}
                     36:
                     37: An application called {\it launcher} is provided to start servers
                     38: and to establish connections as follows.
                     39:
                     40: \begin{enumerate}
                     41: \item A launcher is invoked from a client.
                     42: When the launcher is invoked, the client
                     43: informs the launcher of a port number for TCP/IP connection
                     44: and the name of a server.
                     45: \item The launcher and the client establish a connection with the
                     46: specified port number. One time password may be used to prevent
                     47: launcher spoofing.
                     48: \item The launcher creates a process and executes the server after
                     49: setting the data channel appropriately.
                     50: \end{enumerate}
                     51:
                     52: After finishing the above task as a launcher, the launcher process
                     53: acts as a control server and controls the server process created by
                     54: itself. As to the details of the control server see Section \ref{control}.
                     55:
                     56: As the data channel is used to exchange binary data,
                     57: the byte order conversion is necessary when a
                     58: client and a server have different byte orders. It is determined by
                     59: exchanging the preferable byte order of each peer. If the preference
                     60: does not coincide with each other,
                     61: then the network byte order is used.
                     62: This implies that all servers and clients should be able to
                     63: handle the network byte
                     64: order. Nevertheless it is necessary to negotiate the byte order to
                     65: skip the byte order conversion because its cost is often dominant over
                     66: fast networks.
                     67:
                     68:
                     69: \subsection{Control server}
                     70: \label{control}
                     71: In OpenXM we adopted the following simple and robust method to
                     72: control servers.
                     73:
                     74: An OpenXM server has logically two I/O channels: one for exchanging
                     75: data for computations and the other for controlling computations. The
                     76: control channel is used to send commands to control execution on the
                     77: server. The launcher introduced in Section \ref{launcher}
                     78: is used as a control process. We call such a process a {\it
                     79: control server}. In contrast, we call a server for computation an {\it
                     80: engine}. As the control server and the engine runs on the
                     81: same machine, it is easy to send a signal from the control server.
                     82: A control server is also an
                     83: OpenXM stack machine and it accepts {\tt SM\_control\_*} commands
                     84: to send signals to a server or to terminate a server.
                     85:
                     86: \subsection{Resetting an engine}
                     87:
                     88: A client can send a signal to an engine by using the control channel
                     89: at any time. However, I/O operations are usually buffered,
                     90: which may cause troubles.
                     91: To reset an engine safely the following are required.
                     92:
                     93: \begin{enumerate}
                     94: \item Any OX message must be a synchronized object in the sense of Java.
                     95:
                     96: As an OX message is sent as a combination of several {\tt CMO}
                     97: data, a global exit without sending all may generate broken data.
                     98:
                     99: \item After restarting an engine, a request from a client
                    100: must correctly corresponds to the response from the engine.
                    101:
                    102: An incorrect correspondence occurs if some data remain on the stream
                    103: after restarting an engine.
                    104: \end{enumerate}
                    105:
                    106: {\tt SM\_control\_reset\_connection} is a stack machine command to
                    107: initiate a safe resetting of an engine.
                    108: The control server sends {\tt SIGUSR1} to the engine if it receives
                    109: {\tt SM\_control\_reset\_connection} from the client.
                    110: Under the OpenXM reset protocol, an engine and a client act as follows.
                    111:
                    112: \vskip 2mm
                    113: \noindent
                    114: {\it Client side}
                    115: \begin{enumerate}
                    116: \item After sending {\tt SM\_control\_reset\_connection} to the
                    117: control server, the client enters the resetting state. It discards all {\tt
                    118: OX} messages from the engine until it receives {\tt OX\_SYNC\_BALL}.
                    119: \item After receiving {\tt OX\_SYNC\_BALL} the client sends
                    120: {\tt OX\_SYNC\_BALL} to the engine and returns to the usual state.
                    121: \end{enumerate}
                    122:
                    123: \noindent
                    124: {\it Engine side}
                    125: \begin{enumerate}
                    126: \item
                    127: After receiving {\tt SIGUSR1} from the control server,
                    128: the engine enters the resetting state.
                    129: The engine sends {\tt OX\_SYNC\_BALL} to the client.
                    130: The operation does not block because
                    131: the client is now in the resetting state.
                    132: \item The engine discards all OX messages from the engine until it
                    133: receives {\tt OX\_SYNC\_BALL}. After receiving {\tt OX\_SYNC\_BALL}
                    134: the engine returns to the usual state.
                    135: \end{enumerate}
                    136:
                    137: \begin{figure}[htbp]
                    138: \epsfxsize=8.5cm
                    139: \begin{center}
                    140: \epsffile{reset.eps}
                    141: \end{center}
                    142: \caption{OpenXM reset procedure}
                    143: \label{reset}
                    144: \end{figure}
                    145:
                    146: Figure \ref{reset} illustrates the flow of data.
                    147: {\tt OX\_SYNC\_BALL} is used to mark the end of data remaining in the
                    148: I/O streams. After reading it, it is assured that each stream is empty
                    149: and that the subsequent request from a client correctly
                    150: corresponds to the response from the engine.
                    151: We note that we don't have to associate {\tt OX\_SYNC\_BALL} with
                    152: any special action to be executed by the engine because it is
                    153: assured that the engine is in the resetting state when it has received
                    154: {\tt OX\_SYNC\_BALL}.
                    155:
                    156: \subsection{Debugging facilities}
                    157: Debugging is sometimes very hard for distributed computations.
                    158: We provide two methods to help debugging on X window system:
                    159: 1. the diagnostic messages from the engine are displayed in a {\tt xterm}
                    160: window;
                    161: 2. the engine can pop up a window to input debug commands.
                    162: For example {\tt ox\_asir}, which is
                    163: the OpenXM server of Risa/Asir, can pop up a window to input
                    164: debug commands and the debugging similar to that on usual terminals is possible.
                    165: One can also send {\tt SIGINT} by using {\tt SM\_control\_to\_debug\_mode}
                    166: and it provides a similar functionality to entering the debugging
                    167: mode from a keyboard interruption.