----------------------------------------------------------- CONTENTS 0. Program Files 1. Very Quick Guide 2. Not So Quick Guide 3. Settings in the Tsunami Client 4. Settings in the Tsunami Server 5. Getting Help 6. Realtime Tsunami Notes 7. Special Compile Settings 8. Troubleshooting ----------------------------------------------------------- CHANGES 09 Nov 2007: - added description of heartbeat timeout feature 31 Oct 2006: - some more troubleshooting help (check that no high CPU programs in background, disk is fast enough, etc) 21 Oct 2006: - realtime now has a new UTC time format, recording length can be specified in bytes 17 Oct 2006: - realtime and normal applications now named differently that is, tsunami/tsunamid and rttsunami/rttsunamid ----------------------------------------------------------- 0. Program Files ============= The program files are: Application Command Alternatively call from ----------------------------------------------------- Tsunami server tsunamid ./server/tsunamid Tsunami client tsunami ./client/tsunami Realtime server rttsunamid ./rtserver/rttsunamid Realtime client rttsunami ./rtclient/rttsunami If you have followed COMPILING.txt and the "sudo make install", you can execute the program files from anywhere on the command line. If you omitted the "sudo make install", you'll have to call the programs from within the source code directories. For the general overview on how the Tsunami protocol works, please see the older README.txt 1. Very Quick Guide ================ --Server-- On the server PC, go to the directory where you have the files that should be served, and start the Tsunami server with e.g.: $ cd /where/my/files/are $ tsunamid The server serves files from its current working directory. To allow clients to use the "GET *" command, start the Tsunami server with: $ tsunamid * or $ tsunamid fileToServe1 fileToServe2 ... For the realtime server extensions and its file naming conventions, see the Realtime Tsunami Notes section. --Client-- On the client PC, start the command line client: $ cd /where/to/receive/files $ tsunami The client will store downloaded files into the current working directory. Usual commands in the client are like crude FTP: tsunami> connect server.domain.com tsunami> get filename tsunami> get * 2. Not So Quick Guide ================== Typical usage example where PC1 is the server and PC2 the client. Preparations on PC1: pc1$ cd /home/guest/outfiles/ pc1$ tsunamid * Connecting from PC2: pc2$ cd /home/guest/infiles/ pc2$ tsunami tsunami> set rate 512M rate = 512000000 tsunami> connect myserver.domain.com Connected. tsunami> get nameoffile 00 00 00 368 499 15.59 354.7 0.0 499 0.0 354.7 0.0 0 1 354.7 ... Alternative way of connecting from PC2: - all commands you would enter into tsunami's own command line you may also concatenate onto the command line of your shell, i.e.: pc2$ cd /home/guest/infiles/ pc2$ tsunami set rate 512M connect myserver.domain.com get nameoffile tsunami> 00 00 00 368 499 15.59 354.7 0.0 499 0.0 354.7 0.0 0 1 354.7 ... ... If you use the "get *" command, you can automatically download several files. Note if you use the above shell/command line format of concatenated commands and your shell does globbing, you will have to use "get \*" with a slash. 3. Settings in the Tsunami Client ============ Settings available in the Tsunami Client, and a short description: $ tsunami Tsunami Client for protocol rev 20061025 Revision: v1.1 devel cvsbuild 31 Compiled: Dec 15 2006 14:57:59 tsunami> set server = localhost -- to which currently connected port = 46224 -- TCP control port on server udpport = 46224 -- UDP receiving port on client to use buffer = 20000000 -- size of the ring buffer in RAM in bytes blocksize = 32768 -- how large UDP blocks to use verbose = yes -- give a bit more output transcript = no -- 'yes' to write all screen output to a transcript file, file naming is automatic in "2006-12-15-13-21-41.tsuc" style ip = v4 -- use ip version 'v4' or 'v6' output = line -- output statistics mode is 'line' for scrolling statistics, or 'screen' for a single updating page rate = 650000000 -- the target transfer rate (you may use 'm','M','g','G' so for example '128M' or '1G') error = 7.50% -- maximum error rate to maintain by rate throttling slowdown = 25/24 -- how fast to start throttling the rate speedup = 5/6 -- how fast to recover and move up towards target rate again history = 25% -- weighting of past error rates in the throttling algorithm lossless = yes -- change to 'no' to allow data loss e.g. rate priority transfers losswindow = 1000 msec -- used if lossless='no', sets how old missing data should still be requested again. If too old, data is considered lost. blockdump = no -- 'yes' to write a block bitmap dump to a file, shows packets that were not received as bit value 0, file format is 4 bytes (long) contains number of blocks (bits), followed by number of block count of bits, and two extra bytes that may be ignored passphrase = default -- specify a different non-default passphrase for login to the server 4. Settings in the Tsunami Server ============ $ tsunamid --help Usage: tsunamid [--verbose] [--transcript] [--v6] [--port=n] [--datagram=bytes] [--buffer=bytes] [--hbtimeout=seconds] [filename1 filename2 ...] verbose or v : turns on verbose output mode transcript : turns on transcript mode for statistics recording v6 : operates using IPv6 instead of (not in addition to!) IPv4 port : specifies which TCP port on which to listen to incoming connections secret : specifies the shared secret for the client and server datagram : specifies the desired datagram size (in bytes) buffer : specifies the desired size for UDP socket send buffer (in bytes) hbtimeout : specifies the timeout in seconds for disconnect after client heartbeat lost filenames : list of files to share for downloaded via a client 'GET *' Defaults: ... $ rttsunamid --help Usage: tsunamid [--verbose] [--transcript] [--v6] [--port=n] [--datagram=bytes] [--buffer=bytes] [--hbtimeout=seconds] [--vsibmode=mode] [--vsibskip=skip] [filename1 filename2 ...] [same as above 'tsunamid', plus these:] vsibmode : specifies the VSIB mode to use (see VSIB documentation for modes) vsibskip : a value N other than 0 will skip N samples after every 1 sample Defaults: ... The default 'hbtimeout' after the client heartbeat is lost is 15 seconds. The heartbeat is the packetloss report that the client sends to the server in 350ms or shorter intervals. The 'hbtimeout' behaviour after the first missing client packetloss report is briefly as follows: The server interrupts the transfer after 'hbtimeout' seconds of no client activity. This can happen at any completion %. A backupping realtime server waits until all requested data (100%) has been stored to disk, and then waits for 'hbtimeout'. The same in more detail: - The file server 'tsunamid', or a realtime 'rttsunamid' that has _not_ been instructed to write a local backup copy of VSIB data, will both continue sending new UDP-packetized data while waiting for the next heartbeat. - If no heartbeat is received within the hbtimeout time, the connection is closed. - If an EOF of the TCP connection is encountered, the connection is closed immediately without timeout checks. - A realtime 'rttsunamid' that _has_ been instructed to write a local backup copy of VSIB data will continue to send data, even if there is no feedback from the client. - Even if an EOF is encountered, the connection is not closed. - Only after the server has sent the last block of data (and thus the entire requested VSIB data is complete on the disk), the hbtimeout setting becomes effective. 5. Getting Help ============ Server application, to get a list of command line options use: $ tsunamid --help Client application, to get a list of options use: $ tsunami tsunami> help And to see the current settings in the client (not persistent i.e. not stored between program restarts), use: tsunami> set Settings in the client can be changed using the "set" command which has the format: set [option] [value] You should also read the README.txt. Finally, look at the source code. :-) You can also use the sourceforge project site http://sourceforge.net/projects/tsunami-udp to add a support request or send an email to the developer(s). 6. Realtime Tsunami Notes ====================== The realtime version of the tsunami applications uses the Metshovi VSIB card. This section refers mainly to the realtime versions of the tsunami applications, namely, in the source code tree these are the ./rtserver/tsunamid and ./rtclient/tsunami applications. --VSIB driver-- You'll need a post-01/2006 version of the vsib.c driver. Tsunami tries to access the VSIB card via /dev/vsib. If you try rates of 256Mbps and above, please check that your VSIB driver is loaded with a bigbufsize of at least 100MiB. For example 144MiB often set in the file /etc/modules, and the value is usually sufficient for 512Mbps: # load vsib with 144M vsib bigbufsize=144000000 --Realtime Server-- The realtime server will read data from the VSIB card and send the data over the network. The filename requested by the client in the "get" command tells the realtime server when to start recording. If the name is an EVN formatted time/date string, this will be used as the starting time. Otherwise, recording starts right after the next 1 second rollover. If the file name contains a path i.e. the character "/", the server will locally create a file with that name. Data streamed to the network will then also be saved into that local file. The "EVN" file naming convention with UTC time and optional recording length specified in bytes is as follows: [exp]_[station]_[scan]_[yyyy]-[mm]-[dd]T[hh]:[mm]:[ss].vsi [exp]_[station]_[scan]_[yyyy]-[mm]-[dd]T[hh]:[mm]:[ss]_dl=[bytes].vsi Where Exp: experiment, arbitrary string Station: 2-letter station ID like for example Mh, Jb, Wb, Mc, ... Scan: name of the scan, arbitrary, format often similar to "fr0089" Date: US order, month first Time: UTC time Bytes: dl file length in bytes, length of recording Examples: Metshovi station, begin recording on 17th Oct 2006 at 12:25:30UTC exp01_mh_scan01_2006-10-17T12:25:30.vsi Same as above but also make a local backup on the server into the /i1/ directory (should be a fast RAID0 disk): /i1/exp01_mh_scan01_2006-10-17T12:25:30.vsi Recording length specified to be 1536000000 bytes. exp01_mh_scan01_2006-10-17T12:25:30_dl=1536000000.vsi The default recording length is equal to the amount of bytes in 4 minutes of 512 Mbps. The VSIB recording mode can be specified from the command line options (--vsibmode=x, --vsibskip=x). If you use a version older than v1.1 cvsbuild 32, you are limited to > set blocksize 4096 [or 8192, 16384, 32768] as other settings cause skips in the data. --Realtime Client-- The realtime client will write back data from the network to the VSIB card, i.e. is in "playback" mode. Again, the playback mode is hard-coded but now in ./rtclient/vsibctl.c [Note: realtime client untested after JR wrote it in 2003] To transfer a file from memory/VSIB-in to disk, use the realtime server and normal client i.e.: pc1$ rttsunamid pc2$ tsunami To transfer a file from disk to VSIB-out, use the normal server and realtime client i.e.: pc1$ tsunamid pc2$ rttsunami To transfer a file from VSIB-in to VSIB-out, use the realtime server and realtime client i.e.: pc1$ rttsunamid pc2$ rttsunami (this is not fully tested yet) 7. Special Compile Settings ======================== For real-time server: ./rtserver/io.c //#define MODE_34TH uncomment to enable 3/4th rate transmission, discards upper 4 channels (2 BBCs) For normal client: ./client/protocol.c #define RETX_REQBLOCK_SORTING define to use linear sorting of blocks in the retransmit request, fast, discards duplicates, may improve disk I/O since more contiguous access on server side 8. Troubleshooting =============== * If transfers are really slow, and the error rate is high, check the following on both the client and server: - no other programs should run in the background and use lots of cpu: check with e.g. 'top' - you are really writing to or reading from a fast RAID disk: check with for example 'sudo hdparm -t /dev/md0' or 'sudo hdparm -t /dev/sdb' that the disk where you have the Tsunami data is sufficiently fast compared to Tsunami target Mbps data rate * If no data goes through over the connection, check that no firewalls are blocking the default TCP and UDP port 46224. This can also be tested using the 'netcat' (or 'nc') linux/unix command line tool. pc1$ netcat -u -l -p 46224 pc2$ netcat -u pc1.domain.com 46224 and typing in something into pc2 and hitting should then appear on the pc1 as netcat output, unless there are firewall problems. Firewall settings, server side: TCP 46224 incoming allowed Firewall settings, client side: UDP 46224 incoming allowed * Another possibility is MTU mismatch and no path MTU discovery. Tsunami uses a default block size of 32768 bytes. In the client you could change this to below 10 Mbit LAN MTU using the client console command tsunami> set blocksize 1430 and then reattempt the transfer (commands: connect, get). If this works, then you or your ISP need to fix the network MTU. * If you get "could not create socket on port xxx" errors, this very likely means a zombie tsunamid or tsunami process is running. Try killing that process ('ps axuf', 'kill') or rebooting if it is really stuck. * Try to avoid using the JFS file system (till kernel 2.6.18 or even later) together with Tsunami and RAID, as Tsunami disk thread blocking may hang JFS. In the worst case you loose all data on the file system. Other file systems like EXT3, XFS work just fine, so it is a JFS bug. * If problems persist, contact the developers or submit a bug report at http://sourceforge.net/projects/tsunami-udp * If you use a version older than v1.1 cvsbuild 32, you are limited to > set blocksize 4096 [or 8192, 16384, 32768] as other settings cause skips in the data.