[Dirvish] Dirvish speed and security

[Paul Slootman]

On Mon 26 May 2008, Leopold Palomo-Avellaneda wrote:

> I'm using dirvish 1.2.1-1, from a debian etch. I have created my vaults and 
> configured and running the complete system. I have a nfs mount where I put 
> the backup. It simply works. 

Rsync tries to optimize for network traffic, at a cost of increased
local disk activity / CPU load.  If, however, you fool rsync by using an
NFS mount, rsync thinks it's a local disk and will do the usual high
disk activity which will severely load the NFS network link. It's always
advised to not use rsync on NFS mounts.

Additionally, such network disk boxes are usually quite limited in CPU.
So even if it would be possible to run rsync natively on such a system,
that would probably not perform very well either.

I would recommend using a simple USB / firewire box directly connected
to the system being backupped; that performs at least an order of
magnitude better than an NFS-connected drive, in my experience.

If you're stuck with the situation, you're best off using --whole-file
as option to rsync. That means that rsync doesn't try to use the old
version of the file to optimize the data transferred (reading the old
file over NFS is already causing the data transferred to be 100%, even
without the new data!).  Of course, if you have correctly entered the
local hostname as the client name, then dirvish should automatically use
--whole-file, but I don't see that in the ACTION: line you posted.
What is the output of "uname -n" on that system? Is it "ris"?

If there is a large number of files in the backup, you could also
disable the indexes to prevent a pass with find to get the list of files:

index: none

> Howeber, I have a very poor transfer rate. To create the init backup I need 
> two days to transfer about 90G, and some days, if we have a lot of 

> Total file size: 130790723840 bytes

This shows 130GB, not 90GB.


Paul Slootman