Mailing List Archive: speed up repl

speed up repl

May 20, 2001, 5:26 AM

Post #1 of 6 (1542 views)

Hello,

yesterday i setup two Raid5 boxes with DRBD:
each 2GB RAM, drbd-device (partition) is 140 GB !
connected via x-over cable on 100MBit NICs
drbd version: 0.58 from CVS last Weekend (works very well :)
some lines from messages:
20:12:14 pop2 kernel: drbd: module initialised. Version: 58
20:12:14 pop2 kernel: drbd : vmallocing 4403064 B for bitmap. @fc9aa020
20:12:14 pop2 kernel: drbd0: Connection established.
20:12:14 pop2 kernel: drbd0: size=140898051 KB / blksize=4096 B
20:12:37 pop2 kernel: Checking ReiserFS transaction log (device 2b:00) ...
20:12:37 pop2 kernel: Using r5 hash to sort names
20:12:37 pop2 kernel: ReiserFS version 3.5.29

when i start repl this is really slow, i did some calculation, and i think
my 140 GB will last 3 days for replication (it could last about 4 hours
with my 100MBit )?!

20:13:26 pop2 kernel: drbd0: Synchronisation started blks=7 int=0

after five hours i stopped :

02:23:23 pop2 kernel: drbd0: Synchronisation done. 3444693 blks sent

Is it possible to increase the throughput(blks= ) of drbd ?

>from drbd.conf:
protocol=B
net {
sync-rate=10240
...

--
mfg <o)
Hans Klein /\\
_\_v
/* http://www.net-con.net */

Re: speed up repl [ In reply to ]

philipp.reisner at example

May 21, 2001, 12:13 AM

Post #2 of 6 (1536 views)

Permalink

Hi,

The current replication code should do up to 6MB/sec, (the old
code was running up to 0.6M/sec.).

140GB @ 6MB/sec => 6h

* Hans Klein <hansi.klein@example.com> [010520 14:26]:
> Hello,
>
> yesterday i setup two Raid5 boxes with DRBD:
> each 2GB RAM, drbd-device (partition) is 140 GB !
> connected via x-over cable on 100MBit NICs
> drbd version: 0.58 from CVS last Weekend (works very well :)
> some lines from messages:
> 20:12:14 pop2 kernel: drbd: module initialised. Version: 58
> 20:12:14 pop2 kernel: drbd : vmallocing 4403064 B for bitmap. @fc9aa020
> 20:12:14 pop2 kernel: drbd0: Connection established.
> 20:12:14 pop2 kernel: drbd0: size=140898051 KB / blksize=4096 B
> 20:12:37 pop2 kernel: Checking ReiserFS transaction log (device 2b:00) ...
> 20:12:37 pop2 kernel: Using r5 hash to sort names
> 20:12:37 pop2 kernel: ReiserFS version 3.5.29
>
> when i start repl this is really slow, i did some calculation, and i think
> my 140 GB will last 3 days for replication (it could last about 4 hours
> with my 100MBit )?!
>
> 20:13:26 pop2 kernel: drbd0: Synchronisation started blks=7 int=0

Are you sure thet you are using the new code ?
It should say blks=?? and int=1 (and not 0)

>
> after five hours i stopped :
>
> 02:23:23 pop2 kernel: drbd0: Synchronisation done. 3444693 blks sent

This are about 765 KB/sec (old code?)

>
> Is it possible to increase the throughput(blks= ) of drbd ?
>
> from drbd.conf:
> protocol=B
> net {
> sync-rate=10240
> ...
>
>
-Philipp

Re: speed up repl [ In reply to ]

hansi.klein at example

May 21, 2001, 9:01 AM

Post #3 of 6 (1536 views)

Permalink

At 09:13 21.05.01 +0200, Philipp Reisner wrote:
>Hi,
>
>The current replication code should do up to 6MB/sec, (the old
>code was running up to 0.6M/sec.).
>
>140GB @ 6MB/sec => 6h

my code is from 10.5.
i just downloaded the latest from CVS

--
mfg <o)
Hans Klein /\\
_\_v
/* http://www.net-con.net */

Re: speed up repl [ In reply to ]

philipp.reisner at example

May 21, 2001, 12:46 PM

Post #4 of 6 (1536 views)

Permalink

* Hans Klein <hansi.klein@example.com> [010521 18:01]:
> At 09:13 21.05.01 +0200, Philipp Reisner wrote:
> >Hi,
> >
> >The current replication code should do up to 6MB/sec, (the old
> >code was running up to 0.6M/sec.).
> >
> >140GB @ 6MB/sec => 6h
>
> my code is from 10.5.
> i just downloaded the latest from CVS
>

As "cvs log drbd.c" shows the new syncer was commited on the 14th:
----------------------------
revision 1.99
date: 2001/05/14 10:43:27; author: kde; state: Exp; lines: +213 -145
Syncer rewritten now we can syn with up to 4MByte/sec
----------------------------

-Philipp

Re: speed up repl [ In reply to ]

hansi.klein at example

May 21, 2001, 3:54 PM

Post #5 of 6 (1535 views)

Permalink

At 21:46 21.05.01 +0200, Philipp Reisner wrote:
>* Hans Klein <hansi.klein@example.com> [010521 18:01]:
>> At 09:13 21.05.01 +0200, Philipp Reisner wrote:
>> >Hi,
>> >
>> >The current replication code should do up to 6MB/sec, (the old
>> >code was running up to 0.6M/sec.).
>> >
>> >140GB @ 6MB/sec => 6h
>>
>> my code is from 10.5.
>> i just downloaded the latest from CVS
>>
>
>As "cvs log drbd.c" shows the new syncer was commited on the 14th:
>----------------------------
>revision 1.99
>date: 2001/05/14 10:43:27; author: kde; state: Exp; lines: +213 -145
>Syncer rewritten now we can syn with up to 4MByte/sec
>----------------------------

now it looks better

22:19:03 pop2 kernel: drbd: module initialised. Version: 58
22:19:03 pop2 kernel: drbd : vmallocing 4403064 B for bitmap. @fc90a020
22:19:03 pop2 kernel: drbd0: Connection established.
22:19:03 pop2 kernel: drbd0: size=140898051 KB / blksize=4096 B
22:19:03 pop2 kernel: drbd0: Synchronisation started blks=15 int=1

after two hours:
# cat /proc/drbd
Tue May 22 00:19:00 CEST 2001
0: cs:SyncingAll st:Primary/Secondary ns:30307784 nr:0 dw:4 dr:30307808
gc:1,2,2

what is ns: displaying, blocks, bytes or sectors ?

--
mfg <o)
Hans Klein /\\
_\_v
/* http://www.net-con.net */

Re: speed up repl [ In reply to ]

philipp.reisner at example

May 22, 2001, 5:45 AM

Post #6 of 6 (1545 views)

Permalink

[...]
>
> what is ns: displaying, blocks, bytes or sectors ?
>

It used to be in blocks, but when I rewrote the syncer I decided that
it should be in KB. :) ( Since ext2 changes the block size to 1K
during mounting and uses then the block size that was spezified
at creation time (with the -b option))

Now the numbers are no longer influenced by the block size of the
filesystem, the unit is 1KB.

-philipp