[DRBD-user] Speed Sync seems to be limited on gigabit ethernet

Marcello Lupo mlist at itspecialist.it
Thu Jun 21 09:57:00 CEST 2007

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.

Hi Mike,
yes i'm sure the interface is right... Checking it with ifconfig i see 
tons of gigabytes transferred on the crossover interface and no data on 
the esternal
interface during sync and with tcpdump too.
As i said before i hope there is something to modify in the config to 
reach more speed.

This are the results of syncing 2 different resources (56Gb each) on the 
same 2 machines and disks at the same time:

drbd0: Resync done (total 4428 sec; paused 0 sec; 9992 K/sec)
drbd0: drbd0_worker [31817]: cstate SyncSource --> Connected
drbd1: Resync done (total 5556 sec; paused 0 sec; 10568 K/sec)
drbd1: drbd1_worker [31821]: cstate SyncSource --> Connected

So the connection is able to reach at least 20Mb/sec using drbd. The 
problem seems to be the stability. When i sync only one resource i get
no more than 14/15Mb/sec as average. Instead syncing 2 resources the 
average goes to 20Mb/sec so it seems to be really some limitation
on the drbd resource sync.

Someone else around have a similar config:

2 HP Proliant DL380 with 2 processors Xeon 3Ghz 2 GB ram.
2 disks SCSI Ultra 320 of 146 Gb in raid 1 on each machine  with Smart 
array 6i controller?

As i said when transfer a file between the 2 machines i get 40Mb/sec speed.
There is any performances difference using a dedicated partition to 
store the metadisk data as i'm doing?
I have partition /dev/cciss/c0d0p7 (634 Mb) partition to store metadisk 
data (index 0 for drbd0 and index 1 for drbd1)

Mike Tewner wrote:
> We're configured with a crossover cable on gigabit.
> I just did an invalidate all
> Woops. I hope I did that on the right node :-)
> version: 8.0.3 (api:86/proto:86)
> SVN Revision: 2881 build by root at mx1000-1 , 2007-05-21 09:07:08
>  0: cs:SyncTarget st:Primary/Primary ds:Inconsistent/UpToDate C r---
>     ns:22688 nr:1971135 dw:1986590 dr:63604 al:2 bm:238 lo:258 pe:502 
> ua:256 ap:1
>         [>...................] sync'ed:  1.5% (92552/93958)M
>         finish: 0:17:56 speed: 87,936 (65,428) K/sec
>         resync: used:3/31 hits:122422 misses:184 starving:0 dirty:0 
> changed:184
>         act_log: used:1/127 hits:39136 misses:2 starving:0 dirty:0 
> changed:2
> That's ~65MB/sec, and it hovers around ~62.8MB/sec
> The only relevant lines in the config are:
> common { syncer { rate 100M; } }
> and  perhaps:
> protocol C;
> You can verify that the sync is happening over the correct interface by:
> tcpdump -i <interface> port 7789
> There should be a *LOT* of packets - hit ctrl-c to kill tcpdump

More information about the drbd-user mailing list