Re^4: File::Find won't iterate through CIFS share withouth "dont_use

Reading is definitely my bottleneck. The disk which I read from is working close to its limit, and still the server running the script has some idle CPU time, due to the remote disk's low output. Therefore I need to read from multiple servers (disks) at once.

I believe I do need to iterate, since I don't know which folders might be missing (deleted due to being empty). Remote processes aren't an option, as I have a couple hundreds (windows) servers to search at.

The flow is as follows:

1)Audio files are created on remote servers by some applications.

2)Boss thread searches for filenames on remote servers using File::Find, and enqueues them

3)Worker threads dequeue items and call a conversion application, which converts the file and sends it to a storage (it is never written to the local server's HDD)

4)Upon success, audio file is deleted from remote server

Right now, I need a faster way to search those filenames. I'm trying to grab a few filenames from each remote server, cycling through them (servers) over and over, but I'm not being to feed the queue as fast as it is consumed. So I need to either make it search (and cycle) much faster, or truly make the search multi-threaded.

Comment on Re^4: File::Find won't iterate through CIFS share withouth "dont_use_nlink"

Replies are listed 'Best First'.
Re^5: File::Find won't iterate through CIFS share withouth "dont_use_nlink" by Corion (Patriarch) on Jan 30, 2017 at 15:46 UTC
If your bottleneck is the I/O performance of each server, I would launch a process per-server to scan the directories and potentially (I/O performance permitting) more processes to process each found file from a queue. You seem to have the part of enqueueing and processing the found files in parallel already down, but short of eliminating the network bottleneck by launching processes on the remote servers or by launching local processes for each server, I don't see how your approach could be made faster. You could periodically launch the equivalent of `dir /b /s c:\audiofiles\ > \\centralserver\incoming\audio-%hostname%.l +og` [download] ... or use Perl to do the same. That would somewhat eliminate the network traffic. But if your scanning is bound by the ability of the (windows server) hard disks to deliver the directory entries, that won't help much either. Before you embark on a longer programming spree, you could do a test in the shell to see whether the parallel performance is actually faster than the serial scanning: `TS=$(date '%Y%m%d-%H%M%S') for server in (server01 server02 server03); do ls -1R /mnt/$server/incoming > /tmp/$server-$TS.log & done` [download] This will try to scan three servers in parallel.	[reply] [d/l] [select]

Replies are listed 'Best First'.

Re^5: File::Find won't iterate through CIFS share withouth "dont_use_nlink"
by Corion (Patriarch) on Jan 30, 2017 at 15:46 UTC

If your bottleneck is the I/O performance of each server, I would launch a process per-server to scan the directories and potentially (I/O performance permitting) more processes to process each found file from a queue.

You seem to have the part of enqueueing and processing the found files in parallel already down, but short of eliminating the network bottleneck by launching processes on the remote servers or by launching local processes for each server, I don't see how your approach could be made faster.

You could periodically launch the equivalent of

dir /b /s c:\audiofiles\ > \\centralserver\incoming\audio-%hostname%.l
+og
[download]

... or use Perl to do the same. That would somewhat eliminate the network traffic. But if your scanning is bound by the ability of the (windows server) hard disks to deliver the directory entries, that won't help much either.

Before you embark on a longer programming spree, you could do a test in the shell to see whether the parallel performance is actually faster than the serial scanning:

TS=$(date '%Y%m%d-%H%M%S')
for server in (server01 server02 server03); do
    ls -1R /mnt/$server/incoming > /tmp/$server-$TS.log &
done
[download]

This will try to scan three servers in parallel.

[reply]
[d/l]
[select]