Backup solution suggestions [ggated]

Tom Judge tom at tomjudge.com
Fri Jan 18 11:18:05 PST 2008


Ulrich Spoerlein wrote:
> On Jan 18, 2008 9:11 AM, Johan Ström <johan at stromnet.se> wrote:
>> Your no,barely, bad hell no seems to fit pretty good.. I did some
>> testing during the night with the above (non-production) setup.
>> What I did was doing some rsyncing over the night:
>>
>> while true ; do
>>          echo "`date` Clearing vmail" >> logfile
>>          rm -rf vmail
>>          echo "`date` Starting rsync" >> logfile
>>          rsync -vr /usr/var/vmail . |tee -a logfile
>>          echo "`date` Rsync finished " >> logfile
>> done
>>
>> I started this at ~02.0. The results? A freshly rebooted 6.2 (6.2-
>> RELEASE-p6 FreeBSD 6.2-RELEASE-p6 #0: Fri Jul 27 15:47:50 UTC 2007)
>> box in the morning..
>> [...]
>> What I dont have is a coredump, judging from dmesg -a savecore wasnt
>> even run.. running it now, 5 hours later, didnt find any cores.
>>
>> The other end (7.0 server) wasnt affected at all.
>>
>> Not realy sure what it had been doing, because looking at my
>> bandwidth graphs from the switch, nothing was done at all.. It didnt
>> even go through one iteration of rsync... ~7.5k files/directorys
>> seems to have been transfered, then the log doesnt say more. But
>> according to the BW graph, after ~03.00 no traffic was sent at all...
>>
>> Some known bug with 6.2?
> 
> There was some ggatec problems with TCP and/or sockets, I think they
> have been mostly resolved post-6.2. If you want to pursue this further
> (it *would* be a cool setup, no doubt) I'd suggest three things:
> - Update to 6.3
> - Leave GELI out of the loop for now (only do ggate, with random data perhaps)
> - Build a kernel *without* options PREEMPTION
> 

Hi,

We have 4 production High Avaliability NFS clusters running 
GMirror+GGate+LinuxHA (2 Nodes per cluster) on RELENG_6_2.  This setup 
has proved very stable for us you have to do some tuning though:

/etc/sysctl.conf:
net.inet.tcp.sendspace=1048576
net.inet.tcp.recvspace=1048576
kern.ipc.maxsockbuf=2049152

/boot/loader.conf:
kern.ipc.nmbclusters="32768"


Command line options to ggate[cd]
ggate[dc]_buf_size="1310720"
ggatec_timeout="5"
ggatec_queue_size="2048"

Cluster node uptimes range from 40-160 days with the last reboots being 
caused by power problems not FreeBSD issues.

The problems may be in the tuning or with geli, personally I would leave 
geli out at try with the above configuration.  Then try above with geli 
to see what the problem is.

Tom









More information about the freebsd-stable mailing list