[Box Backup] Change in network usage characteristics in 0.08?
Sun, 28 Nov 2004 15:40:53 -0800
On 11/27/04 9:10 AM, Ben Summers wrote:
> There were no changes in protocol behaviour in 0.08.
> If you use lazy mode, the daemon is designed to send a trickle of
> backups throughout the day. So I would expect a connection to the
> server for quite a lot of the day if the clients have extensive data
> modifications to upload.
Agreed. However, I'm seeing this trickle of data 24x7. No let-up. I know
that the folks that are using the machine aren't changing files at 2 in
the morning. They go home at 5pm, so the latest I should see is about
11pm or midnight for any substantial data transfer.
> Regarding the asymmetry over the data transfer, it does seem a little
> odd. Obviously you're confident that the measuring system is correct,
> so turn on ExtendedLogging and take a look at the logs and see
> exactly what is being sent.
I'm using MRTG, which is a robust bandwidth monitoring tool.
> If a directory is modified on the client, it will request a list of
> that directory from the server until it has uploaded the files. This
> might mean it's requested once an hour with the default settings, for
> up to six hours. Might this be it?
> Or if the client is a fileserver, and one of it's clients has a clock
> which is wildly out of sync, it might be downloading the directories
> all the time. Check the client logs for warnings about huge offsets,
> but, say, a 12 hour offset won't provoke the warning but continual
> filesystem write activity by that fileserver client will result in
> more queries to the server.
> But as usual, a look at the logs will give more clarity over what's
I looked at the logs, and here's what I've found: The two wayward
clients are connecting every 5 minutes and 10 seconds (+/- 3 seconds),
rather than every hour as their config files dictate.
Your explanation about the higher transmission rates makes sense, if the
clients are constantly requesting the directory information from the
server. The connection takes between 2 and 4 minutes, so there is always
at least one client connected to the server. Hence the constant
I won't have access to the client machines until tomorrow (they are at 2
different companies that won't be open until Monday), but I will look at
the Windows event logs for them to see if I can glean any more
information from them.
My suspicion is that one of two things is happening:
1. There is a bug in the 0.08 cygwin client, which causes this behavior.
My bandwidth use changed after I installed the 0.08 cygwin client. I
installed the 0.08h win32 client on a Windows machine here, and observed
no problems like what I'm seeing with the cygwin client. The cygwin
client on this machine had done the 'every-5-minute' thing before that
2. The new versions of the client somehow has permission problems with
the /var/bbackupd directory. I will be checking into that tomorrow. If
bbackupd is unable to touch the 'last_sync_start' file, I can see how
this can happen. last_sync_start never gets a more recent timestamp, and
bbackupd will continually try to touch the file, and re-connect.
I'm going to be switching to the Win32 clients anyway, so this may be a
moot point for me (if it is cygwin-related), but I wanted to make sure
that something wasn't being missed.
> On 27 Nov 2004, at 06:02, Per Thomsen wrote:
>> I have noticed in the last couple of weeks (after I installed 0.08)
>> that traffic with my box-clients has changed its behaviour: Now I
>> have a constant traffic of about 30-50 kbps outbound to the client
>> machines. Is this expected behaviour?
>> I have 3 clients that run over the connection in question. The
>> traffic seems to be steady at about the same rate all the time,
>> according to MRTG.
Per Reedtz Thomsen | Reedtz Consulting, LLC | F: 209 883 4119
V: 209 883 4102 | firstname.lastname@example.org | C: 209 996 9561
GPG ID: 1209784F | Yahoo! Chat: pthomsen | AIM: pthomsen