| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 2234.1 | wait for some iea | DEKVC::JINSUNPARK | | Mon May 19 1997 13:51 | 18 |
| Hello,
I found some error in ER* file as below and this error repeats.
Network-Error:
/usr/op/servers/csd at Tue May 6 11:59:49.903: couldn't send data to
host"" -
error: 32, Broken pipe
position of error not known
dynamic function trace:
==> CSUuserSelNext ==>CSUuserSelNext ==>CSUuserSelNext ==>
==> CSUuserSelNext ==>CSUuserSelNext ==>CSUuserSelNext ==>
==> CSUuserSelNext ==>CSUuserSelNext ==>CSUuserSelNext ==>
|
| 2234.2 | details on internal mail | UTROP1::16.197.208.129::VISSER_J | Joop Visser @ UTO | Tue May 20 1997 20:23 | 97 |
| Re .0
The Mail function incorporates functions Quickmail, To User, To Org. Unit,
Quick Reply and Send. The details of operation of these requests are described
to this document and is based on LinkWorks release 3.08 and higher. Where it
is applicable differences with earlier releases are mentioned.
Functionality
With the Quickmail function one can forward a message to another user. This
function does not use the mail queing facilities which means that if this
request is completed the addressed user will have the message stored into the
MailInBox. One envelope object is created containing the message and subject.
The Send function has three sub modes : Forward, Copy and Shared.
The forward mode does not make use of the queueing either. One user is
addressed and the attached Linkworks object is transferred to this user and
stored in the created envelope object. Again, if the request is completed the
addressed user has in the MailInBox an envelope containing the selected
object.
Mode Copy and shared make use of the queueing facility. The mailjob will be
processed where the priority of mailjobs are taken into account. The
processing is done by pdmbgjob which is activated by pdmbg which runs on each
LinkWorks server. Process pdmbgjob will be active until there is no mailjob to
be processed on that server. When at a later time new mail is send a new
pdmbgjob is started by pdmbg.
The 3.07 implementation:
The processing is rather expensive on CPU time and database usage. Therefor
not all mail is processed at once but selected portions are processed and then
paused for delay(*) seconds until the next selected portion will be processed.
One such processing slot is refered to as mail cycle. If the delay is set to
zero it could lead to "cell not available" message on actions done by the end
user. This is caused because a connect to a free celld deamon request has
timed out due to increased usage of cell deamon processes. Other servers in
the LinkWorks domain can activate actions related to mail.
(*)(delay defined in /usr/op/servers/startpdm.sh by pdmbg option -d)
The 3.08 implementation:
The delay default has been changed to 0 (from 30) seconds. The background
process pdmbgjob has got a higher nice value resulting in a lower priority
than the LinkWorks processes. From OS point of view it becomes a background
process. This must improve the overall performance.
Database mailjob administration
A mailjob is administrated in the communication cell. Each mailjob is
described in table mailjob by a row with methode id 230(copy) or 232(share),
describing the envelope object and a second row to indicate end of mailjob
(method id 266). In table mailjobadr are stored the addressees for a mailjob,
one row per addressee. An end_of_mailjob also has 1 row in the table
mailjobadr.
Database mail processing
One mail cycle can process up to 8 addressees belonging to the same mailjob.
The end of mailjob takes also one mail cycle. First the method 23x entries are
selected and sorted on priority and mailjob id. From the first mailjob in the
queue 8 addressee are taken and the mail for these selected addressees are
processed. If there is 1 addressee only this addressee is handled.
NOTE:
In releases prior to 3.08 always 1 addressee is handled per mail cycle.
If there are no method 23x entries the method 266 entries are handled. Per
mail cycle 1 row. If there is a mail send this new mail gets processed first
as the envelop gets method 23x and this has priority above the 266 methods.
If there are no more rows in the table mailjob the next cell is searched for
entries. This has the effect that if there are more cells on 1 server and if
the communication cell selection is scattered over all cells those users
having their communication cell on the first cell have priority over those
users having the communication cell on the other cells with a higher
identification. It was the intention in design that there was only 1
communication cell per server.
At the end of may97 an update of libpdm.so will be made available which will
schedule between all local cells on the server. If a cell has no mail the next
cell is searched for mail. If there is no mail in all local cells the pdmbgjob
exits.
Performance and capacity details
Release 3.07
In release 3.07 a delay of 30 seconds is taken as default. The waiting delay
is mentioned under throughput in the table mentioned below.
Case definitions with delay of 30 seconds(default):
1. Mail copy to 600 addressees
2. Mail copy to 1 addressee
3. 600 mailjobs person to person copy mode.
Case Release Mail cycles Throughput
1 3.07 600 + 1 = 601 5 hours and 30 seconds
2 3.07 1 + 1 = 2 1 minute
3 3.07 600 + 600 = 1200 10 hours
Remark to case 3: In 5 hours all addressed users will have received the mail
but the server will be busy processing the method 266 (end_of_mailjob).
Release 3.08
From 3.08 onwards the delay is set to zero seconds giving the throughput time
result a total of CPU and database usage.
Joop Visser
|
| 2234.3 | takes long time than expected time | DEKVC::JINSUNPARK | | Thu May 22 1997 16:35 | 209 |
|
Hello, Joop
I'm very great to hear from you and it helped me.
But still i have the problem.
Our customer didn't accept mail throughout time as below result time.
I'd like to know if below test data is a natual result or we have
something wrong.
I tested many times as changing delay time when no login users
and got the result table as below.
Why does it take long time than our expected time ?
There was no cpu and disk i/o, memory load.
Active login user number are 200-300 and user number registered are
about 7000.
<<< The result table >>>
delay time(d) addressees num. Throught(min) expected Throught
================================================================
3 320 34 16
1 320 14 6
1 800 38 14
0 320 7 5
0 800 37 7
Customer environment.
S/W : digital unix 3.2D
LINKWORKS 3.0.8
ORACLE 7.2
TOTAL CELL NUMBER : 3
H/W : 2100 * 3 (ALPHA)
==========================================================
| | |
2100(HDMS01) 2100(HDMS02) 2100(HDMS03)
( CELLA ) ( CELLB ) ( CELLC )
cltdf.cfg file contents
opcs[0] :UNIX::/usr/op/servers/csd.sck
minserver=4
maxserver=20
startminservers=true
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
opcell[1] :UNIX::/usr/op/servers/celld1.sck
minserver=8
maxserver=35
startminservers=true
minload=30
maxload=80
maxclientwait=5
minserveridle=20
connecttimeout=5
responsetimeout=5
sensitivity=50
opdcs[1] :UNIX::/usr/op/servers/dcs.sck
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
tdfctl[1] :UNIX::/usr/op/servers/tdfctld.sck
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
opana[1] :UNIX::/usr/op/servers/analysed.sck
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
opcell[2] :inet:hdms02.hanjung.co.kr:17001
minserver=8
maxserver=35
startminservers=true
minload=30
maxload=80
maxclientwait=5
minserveridle=20
connecttimeout=5
responsetimeout=5
sensitivity=50
opdcs[2] :inet:hdms02.hanjung.co.kr:17001
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
tdfctl[2] :inet:hdms02.hanjung.co.kr:17001
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
opna[2] :inet:hdms02.hanjung.co.kr:17001
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
opcell[3] :inet:hdms03.hanjung.co.kr:17001
minserver=8
maxserver=35
startminservers=true
minload=30
maxload=80
maxclientwait=5
minserveridle=20
connecttimeout=5
responsetimeout=5
sensitivity=50
opdcs[3] :inet:hdms03.hanjung.co.kr:17001
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
tdfctl[3] :inet:hdms03.hanjung.co.kr:17001
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
opana[3] :inet:hdms03.hanjung.co.kr:17001
minserver=1
maxserver=3
startminservers=false
minload=45
maxload=80
maxclientwait=2
minserveridle=10
connecttimeout=15
responsetimeout=5
sensitivity=50
|
| 2234.4 | experienced performance | UTROP1::16.197.208.129::VISSER_J | Joop Visser @ UTO | Fri May 23 1997 16:12 | 17 |
| Hello Jin Sun,
From the delay time 0 throughput figures the average processing time for 1
addressee is 1,3 seconds (320) and 2,77 seconds (800) which is in the scope of
expienced performance.
On a production system it is difficult to measure performance as the database
access can not be controlled. Tables can grow as well which will have
different start conditions per test, especially with mail as the mail tables
grow.
For person to person mail you can overcome this problem by creating a share of
the object and then forward this share to the other person.
Regards,
Joop Visser
|