Aw: Re: Re: [PATCH] vnstat: Update to 1.18

[Bernhard Bitsch <Bernhard.Bitsch@gmx.de>]

[-- Attachment #1: Type: text/plain, Size: 4139 bytes --]



> Gesendet: Freitag, 23. März 2018 um 16:54 Uhr
> Von: "Michael Tremer" <michael.tremer(a)ipfire.org>
> An: "Bernhard Bitsch" <Bernhard.Bitsch(a)gmx.de>, "IPFire Development" <development(a)lists.ipfire.org>
> Betreff: Re: Aw: Re: [PATCH] vnstat: Update to 1.18
>
> On Fri, 2018-03-23 at 13:39 +0100, Bernhard Bitsch wrote:
> > > Gesendet: Mittwoch, 21. März 2018 um 20:30 Uhr
> > > Von: "Bernhard Bitsch" <Bernhard.Bitsch(a)gmx.de>
> > > An: "Matthias Fischer" <matthias.fischer(a)ipfire.org>
> > > Cc: "IPFire: Development-List" <development(a)lists.ipfire.org>
> > > Betreff: Aw: Re: [PATCH] vnstat: Update to 1.18
> > > 
> > > 
> > > > Gesendet: Mittwoch, 21. März 2018 um 18:38 Uhr
> > > > Von: "Matthias Fischer" <matthias.fischer(a)ipfire.org> 
> > > > An: "Michael Tremer" <michael.tremer(a)ipfire.org>, "IPFire: Development-
> > > > List" <development(a)lists.ipfire.org>
> > > > Betreff: Re: [PATCH] vnstat: Update to 1.18
> > > > 
> > > > > 
> > > > > P.S. I did not appreciate that comment of that forum user about this
> > > > > bug. Bugs
> > > > > happen. If he doesn't like it, he can take lots of €€€€ into his hand
> > > > > and pay
> > > > > other people to get their bugs.
> > > > 
> > > > Me too. But I tried to be patient, took a deep breath and tried to keep
> > > > calm and ~professional. Wasn't easy. Besides, I don't think this is
> > > > really a "bug". 'vnstat' can handle his speed and download rates with no
> > > > problems. It just happened that they were a bit too much for the
> > > > standard config - in *his* case.
> > > > 
> > > > Best,
> > > > Matthias
> > > > 
> > > > 
> > > 
> > > Hi,
> > > 
> > > just my opinion ( I've investigated now a bit about this "problem" ;) )
> > > The "solution" isn't really that. The changed parameter is only used by
> > > vnstatd, which isn't used in IPFire. I can't imagine what did the remedy.
> > > We don't know exactly his configuration. Therefore I gave him some hints to
> > > help clarifying the case.
> > > If he doesn't answer, just forget his unqualified comment.
> > > 
> > > Best,
> > > Bernhard
> > > 
> > 
> > News about the vnstat problem.
> > Intensive PMs with Zonediver showed that the effect is caused by 32 bit byte
> > counters of the NIC. ( vnstat reads /proc/net/dev )
> > He changed the fcrontab period for makegraphs to 3 minutes now. Amount of
> > bytes received in this period is less than 2^32.
> > 
> > Remain two questions: 
> > - Is this a normal behaviour for the combination of i210-T1 NIC and associated
> > driver or is this specific to his system?
> 
> I cannot speak for that NIC because I do not use that anywhere. Surprised to
> hear this the first time since the i210 is also on the APU boards and it
> actually is a stripped down version of a bigger Intel NIC.
> 
> > - Can we use the vnstatd daemon? This would allow to increase the vnstat
> > update frequency independent from makegraphs.
> 
> I suppose we could. Arne has built this into the distribution and could probably
> comment more. If I remember correctly, the daemon wasn't available when we
> adopted vnstat.
> 
That's right.
> But even collecting once a minute wouldn't be enough because with max.
> theoretical throughput of a Gigabit NIC, 2^32 bytes are transferred in 32
> seconds. And if this is only a problem with that specific NIC, I actually
> wouldn't want to collect every 30 second on all systems. Wastes a lot of
> resources and costs a lot of write cycles on flash.
> 
> > I've found a thumb rule for the maximal update period for a given bandwidth.
> > I've documented this in the forum thread.
> 
> > 
> > Best,
> > Bernhard
> > 
> 
> -Michael
> 

The default value of vnstatd are 30 seconds for UpdateInterval (short enough for 2^32 bytes on 1 GBit/s ), writes to disk is ruled by parameter SaveInterval ( default 5 minutes, which is our period from cron ).
Thus vnstatd should give a better resolution ( for some systems without information leaks! ) with similiar load as the cron based solution implemented at the moment.

-Bernhard