Prev: Call for Paper The International Journal of Computer Science (IJCS)
Next: python & perl bigint difference
From: #! /shell/nerd on 2 Mar 2010 12:41
Hi,

I have an interesting performance problem - I'm creating a file /tmp/
out in a server & putting the JVM startup progress in it. The typical
content of this file could be -
JVM01 .....STARTED
And I have 10 such JVMs in a machine so they all simultaneously try to
write to this file at one instance. Now I found that Atleast one is
missed during the simulataneous write at the beginning e.g. JVM08
might not be able to get a chance to put it's entry in the file. Now
the work around I have is to check the entry for that JVM in this file
in a look (say 10 times) & put the entry if not found. Is there a
better way of doing it?
From: pk on 2 Mar 2010 12:42
#! /shell/nerd wrote:

> I have an interesting performance problem - I'm creating a file /tmp/
> out in a server & putting the JVM startup progress in it. The typical
> content of this file could be -
> JVM01 .....STARTED
> And I have 10 such JVMs in a machine so they all simultaneously try to
> write to this file at one instance. Now I found that Atleast one is
> missed during the simulataneous write at the beginning e.g. JVM08
> might not be able to get a chance to put it's entry in the file. Now
> the work around I have is to check the entry for that JVM in this file
> in a look (say 10 times) & put the entry if not found. Is there a
> better way of doing it?

How are you writing to the file? If you're independently running many
instances of code like ".... > /tmp/out", then I'm not surprised that some
information is missing, although in this case the outcome would be different
from what you describe (IIUC).
From: #! /shell/nerd on 2 Mar 2010 12:54
On Mar 2, 11:42 am, pk <p...(a)pk.invalid> wrote:
> #! /shell/nerd wrote:
> > I have an interesting performance problem - I'm creating a file /tmp/
> > out in a server & putting the JVM startup progress in it. The typical
> > content of this file could be -
> > JVM01  .....STARTED
> > And I have 10 such JVMs in a machine so they all simultaneously try to
> > write to this file at one instance. Now I found that Atleast one is
> > missed during the simulataneous write at the beginning e.g. JVM08
> > might not be able to get a chance to put it's entry in the file. Now
> > the work around I have is to check the entry for that JVM in this file
> > in a look (say 10 times) & put the entry if not found. Is there a
> > better way of doing it?
>
> How are you writing to the file? If you're independently running many
> instances of code like ".... > /tmp/out", then I'm not surprised that some
> information is missing, although in this case the outcome would be different
> from what you describe (IIUC).

Yeah forgot to mention that ..
Before starting any JVM, In it's own process I'm doing printf "JVMXX
\n" >> /tmp/out
So each JVM writes to the file first & then continues with the start.
From: #! /shell/nerd on 2 Mar 2010 21:32
On Mar 2, 11:54 am, "#! /shell/nerd" <vikas...(a)gmail.com> wrote:
> On Mar 2, 11:42 am, pk <p...(a)pk.invalid> wrote:
>
>
>
> > #! /shell/nerd wrote:
> > > I have an interesting performance problem - I'm creating a file /tmp/
> > > out in a server & putting the JVM startup progress in it. The typical
> > > content of this file could be -
> > > JVM01  .....STARTED
> > > And I have 10 such JVMs in a machine so they all simultaneously try to
> > > write to this file at one instance. Now I found that Atleast one is
> > > missed during the simulataneous write at the beginning e.g. JVM08
> > > might not be able to get a chance to put it's entry in the file. Now
> > > the work around I have is to check the entry for that JVM in this file
> > > in a look (say 10 times) & put the entry if not found. Is there a
> > > better way of doing it?
>
> > How are you writing to the file? If you're independently running many
> > instances of code like ".... > /tmp/out", then I'm not surprised that some
> > information is missing, although in this case the outcome would be different
> > from what you describe (IIUC).
>
> Yeah forgot to mention that ..
> Before starting any JVM, In it's own process I'm doing printf "JVMXX
> \n" >> /tmp/out
> So each JVM writes to the file first & then continues with the start.

I'd appreciate if somebody could suggest anything !!
From: Geoff Clare on 3 Mar 2010 08:39
#! /shell/nerd wrote:

>> > > And I have 10 such JVMs in a machine so they all simultaneously try to
>> > > write to this file at one instance. Now I found that Atleast one is
>> > > missed during the simulataneous write at the beginning e.g. JVM08
>> > > might not be able to get a chance to put it's entry in the file.

>> Before starting any JVM, In it's own process I'm doing printf "JVMXX
>> \n" >> /tmp/out
>> So each JVM writes to the file first & then continues with the start.
>
> I'd appreciate if somebody could suggest anything !!

What shell are you executing that printf in? If it is not a
POSIX-conforming shell then it's possible that it is implementing ">>"
by just seeking to the end of the file instead of setting O_APPEND on
the file descriptor.

--
Geoff Clare <netnews(a)gclare.org.uk>


 |  Next  |  Last
Pages: 1 2
Prev: Call for Paper The International Journal of Computer Science (IJCS)
Next: python & perl bigint difference