| View previous topic :: View next topic |
| Author |
Message |
loopx
Advocate
Joined: 01 Apr 2005
Posts: 2787
Location: Belgium / Liège
|
 Posted: Wed Apr 21, 2010 4:06 pm Post subject: [DRBD] Much network transfert than copied data |
 |
|
Hello,
I'm trying DRBD and making some test. I found a strange behavior using these configuration :
- ext3 on DRBD mounted with "sync" option
- DRBD protocol C
- DRBD configuration :
| Code: |
# last update by loopx on 20100414 :
# - update the configuration to be easier with version 8.3
# - creation of this config
#################################################################
global {
#don't send statistics through internet ...
usage-count no;
}
common {
protocol C;
syncer {
#10Mbyte per second
rate 10M;
#use compression for bitmap exchange
use-rle;
}
}
resource r0 {
device /dev/drbd0;
meta-disk internal;
#use "none" for write-after-write (because, we've got battery for our server)
disk {
no-disk-barrier;
no-disk-flushes;
no-disk-drain;
}
on wk02lhe.<domain> {
address <ip1>:7789;
disk /dev/loop0;
}
on wk01lhe.<domain> {
address <ip2>:7789;
disk /dev/loop0;
}
}
|
In fact, I'm trying to have the faster replication as possible ; I know there is lost of performance, but I just want to try.
The first test was made without any synced filesystem ... all was fine : about 10Mb/s through the network.
But now, with "synced" filesystem, I got about 5Mb/s through the network ... but this is not all ... Currently, the transfert over the network is 14Gb ... and when I have a look on the DRBD mounted partition (where I'm currently copying a 10Gb file), I only see a file of 1.7Gb ....
So, why is only 1.7Gb copied on the DRBD (and, probably, replicated on the other node) and the network show me than there is more than 14Gb transfered ? How this is possible since the file to transfert is only 9.6Gb  ??????????
I have a look on the CPU IO wait and ... it's about 50% of IO wait ... may be it's because of that wait time ?
I really not understand ... All other test (without "sync" on filesystem) completed the replication when the network has sent 10Gb ... not over ...
Any idea of what's happening ?
_________________
Mon MediaWiki perso : http://pix-mania.dyndns.org |
|
| Back to top |
|
 |
richard.scott
Veteran
Joined: 19 May 2003
Posts: 1497
Location: Oxfordshire, UK
|
| Posted: Sun Apr 25, 2010 5:30 pm Post subject: |
|
|
The 'sync' flag will slow down your throughput lots as it forces the data to be flushed to the disk each time something is written.
As your doing this on a mounted drbd device you'll need to get this feedback from the other node before your node will confirm the write.
This will a) increase your bandwidth usage (as you'll be sending data, and getting confirmation back) and b) slow down your write speed.
I would not use the sync option, but use the option in DRBD instead. |
|
| Back to top |
|
|
loopx
Advocate
Joined: 01 Apr 2005
Posts: 2787
Location: Belgium / Liège
|
| Posted: Thu May 06, 2010 9:03 am Post subject: |
|
|
| richard.scott wrote: | The 'sync' flag will slow down your throughput lots as it forces the data to be flushed to the disk each time something is written.
As your doing this on a mounted drbd device you'll need to get this feedback from the other node before your node will confirm the write.
This will a) increase your bandwidth usage (as you'll be sending data, and getting confirmation back) and b) slow down your write speed.
I would not use the sync option, but use the option in DRBD instead. |
Thanks you, I will not use EXT3 with "sync" option. Now, I'm looking for DRBD + GFS to make a little cluster R/W and RO 
Should be better 
_________________
Mon MediaWiki perso : http://pix-mania.dyndns.org |
|
| Back to top |
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
|
Kernel & Hardware
