Le Wednesday 30 March 2011 15:41:29 Alan Brown, vous avez écrit :
> Laurent HENRY wrote:
> > I let bacula a few days with attribute spooling.
> > Things are not really different but it is a nice feature anyway.
> >
> > I am still not beyond 100Mb/s on my 1Gb/s network.
>
> You should be able to hit 300-400Mb/s assuming you are spooling to a
> mechanical hard drive or cheap SSD. Faster than that requires an array
> of disks setup for striping.
Well, i am spooling to a server with Nearline SAS disks (10k) on RAID1/RAID 5
and a huge iSCSI disk area.
I already thought the bottleneck could be the client, which i don't know how
to improve.
What makes me trying to do something on the server is, with other backups with
differents jobs are not above 100Mb/s either
On this 2nd simple configuration:
Client with a dedicated Gb interface on a private VLAN
Server with a dedicated Gb interface on the same private vlan.
Both directly connected through a Cisco catalyst 6500
Backups on a LTO-4 tape drive directly connected via SAS on the bacula server.
Trying a iperf between both machines gives me a Gb speed as expected:
# /usr/bin/iperf -s -i 2
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 85.3 KByte (default)
------------------------------------------------------------
[ ID] Interval Transfer Bandwidth
[ 4] 0.0- 2.0 sec 224 MBytes 942 Mbits/sec
[ 4] 2.0- 4.0 sec 224 MBytes 941 Mbits/sec
[ 4] 4.0- 6.0 sec 224 MBytes 941 Mbits/sec
[ 4] 6.0- 8.0 sec 224 MBytes 941 Mbits/sec
[ 4] 8.0-10.0 sec 224 MBytes 941 Mbits/sec
[ 4] 0.0-10.0 sec 1.10 GBytes 941 Mbits/sec
------------------------------------------------------------
During last backup i had few peaks at 110 Mb/s, nothing more. Average trafic
is more about 30Mb/s
>
> > Did someone tried to tweak some network parameters (On linux Debian) ?
> > I think about Jumbo frames, but because it is a vlan-wide parameter and
> > bacula is not the same on this network, it is a little complicated to
> > deploy.
>
> I have some settings but they won't gain much until other areas are
> addressed.
>
> Before you attempt to do anything, you will need to benchmarking network
> speeds and speeds of simple file transfers, etc. You need to get a
> handle on how fast your network and disk susbsystems are before you can
> see how fast Bacula can run.
>
> In my experience throughputs are far more limited by client disk
> speed/loading (busy clients can be very slow to provide data), server
> disk speeds/loading and finally by network limitations - in that order.
> Tuning networking when the bottlenecks are elsewhere will give few
> noticeable improvements until those other areas are quantified and
> addressed if needed.
>
> I can easily sustain 1Gb/s on terabyte-scale backups without jumbo
> packets, but it took a bit of tuning of the client's disk handling to
> achieve it and I have hardware arrays with 96 drives onboard apiece. The
> tuning was mainly intended to improve its performance as a fileserver
> and came at the expense of its usefulness as an interactive machine
> (it's a dedicated fileserver) - there's no such thing as a free lunch.
--
Laurent HENRY
Administrateur Systèmes & Réseaux
Responsable du CRI
RSSI
EHESS - CRI
190 Av de France
75013 Paris
Secrétariat du CRI: 01 49 54 23 08
Tel: 01 49 54 23 61
------------------------------------------------------------------------------
Create and publish websites with WebMatrix
Use the most popular FREE web apps or write code yourself;
WebMatrix provides all the features you need to develop and
publish your website. http://p.sf.net/sfu/ms-webmatrix-sf
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users
|