Re: Speed and regression

On 2024-09-27 05:29, Eric Lavarde wrote:
> Hi,
> 
> On 27/09/2024 07:09, Duffer wrote:
>> On 2024-09-25 22:30, Duffer wrote:
>>> Hi.
>>> 
>>> I'm using rdiff-backup 2.2.6 for a large amount of data. All traffic 
>>> is encrypted via stunnel.
>>> 
>>> I have two concerns:
>>> 
>>> 1. Speed
>>> 
>>> With SSH, the speed I've seen is 25-50 Mbps, probably averaging 30-40 
>>> Mbps. iperf3 gets around 500 Mbps between my local system 
>>> (destination) and the remote system (source). The SSH speed seems 
>>> excessively slow to me - maybe 6-8% of the measured bandwidth.
>>> 
>>> I switched to mapping the data source via NFS. Now I'm seeing full 
>>> speed for the transfers.
>>> 
>>> Why is SSH so incredibly slow? Is that expected? I'm okay using NFS 
>>> but I liked the convenience of SSH.
> 
> Well, I can only guess that it's the pickling and unpickling necessary 
> over the network. I'm working on improving performance but it takes 
> time. I first need to refactor and simplify the code. It could also be 
> worsened due to a weak CPU on one end or the other. Many small files 
> also increase a lot the overhead.

Good to know. I think SSH has a lot of overhead, anyway. It's usually 
far from the fastest protocol. There's at least one project attempting 
to improve performance: https://www.psc.edu/hpn-ssh-home/

>>> 2. Regression
>>> 
>>> As noted, my backup was very slow so I stopped it (CTRL-C). The next 
>>> time I ran the backup, it went into "regress" mode, so I lost all the 
>>> progress. Is there no way to avoid this and resume somehow? I 
>>> couldn't find anything in the FAQ.
> 
> No, but there is already an issue for this. Again, takes time...

Thanks. Yes, it's definitely something you want to get right before 
deploying it.

>> The regress didn't take long at all but it's now been running over 10 
>> hours and I've seen little progress. Disk usage hasn't increased at 
>> all. Is this expected? I thought after the regress it would start 
>> downloading, like it did before. I can't tell what it's doing but I 
>> seem to see it reading a lot from the disk. I used "-v 3" but maybe I 
>> should have opted for more output.
> 
> -v3 is the default value, I generally use -v5 to see progress (you can 
> configure differently logfile and console output BTW).
> Difficult to say what's happening from here, but did you check if files 
> are appearing in your backup repo, especially under rdiff-backup-data?
> If everything breaks, assuming Linux, you can try to guess what it's 
> doing using strace.
> 
> KR, Eric

I'm seeing continuous creation of files like rdiff-backup.tmp.217452 in 
the backup path. There are also lots of files under rdiff-backup-data, 
including two one file and one directory from today.

It's been running 18 hours, so I hate to break it again but I really 
wish I could figure out what it's doing. One of the backup directories 
(where I see the tmp files) keeps growing and shrinking.