[rrd-developers] two problem about RRD UPDATE

[Florian Forster]

Hi,

On Sat, Mar 14, 2009 at 12:01:42PM +0800, lily wrote:
>     Recently I use the rrdtool to create about 12 thousand rrd-file, and
> update them in 30s .

the important metric here is updates per second. Updating 12k RRD files
every 30 seconds qeuals 400 updates per second. In my experience, you
can expect about 120-150 updates per second from decent disks. If you
can stay in memory, you can do significantly more updated of course..

> [...], and the FS is ext2

> [...], and do the vm Optimizations. But it seems not work. The IO
> await list is sometime very high.

You did mount with `noatime' and `nodiratime', right? While this doesn't
perform miracles, it'll certainly increase performance some percent..

>        Device:    rrqm/s wrqm/s   r/s   w/s  rsec/s  wsec/s    rkB/s   wkB/s avgrq-sz avgqu-sz   await  svctm  %util
>       cciss/c0d0   0.00   0.00  0.00 112.44    0.00  907.77     0.00  453.89     8.07   112.99 1001.39   9.23 103.73

450 kBytes/s seems a bit low for a RAID-10 (or RAID-0+1? I keep
forgetting which one is the one with the better write performance ;)
setup. With 15k disks you should be able to get as high as
1.5 - 2.0 MBytes/s. Can you tell us some more about your system? How
much RAM do you have? What kind of disks? What are the precise options
you used when mounting the file system? This kind of stuff..

>     The second is sometime when I update the files , there is an error :
>          illegal attempt to update using time 1236857494 when last update
> time is 1236857494 (minimum one second step)
>      but I use the timestamp is 1236886011, not the 1236857494. And I use
> the rrdtool last to see the last update time of the file , I find the last
> update time is 1236857468, also not 1236857494. I have debug a lot of time ,
> but have no idea about it .

I haven't observed this problem yet, but due to misconfiguration by
users I see this message a lot (when it is legitimate). I see other
error messages on two machines that are under heavy load though. Such
messages include:

 - closing rrd
 - reading the cookie off $FILE faild [sic]
 - is not an RRD file

When things go really bad, I get "Unknown CF ''" followed by a SIGFPE.

I suspect that the high load on the machines leads to some incomplete
reads/write which cause these messages/problems. Unfortunately I don't
have debugging symbols on these machines and no way of getting some onto
there, so haven't been able to track these problems down yet :/

Regards,
-octo
-- 
Florian octo Forster
Hacker in training
GnuPG: 0x91523C3D
http://verplant.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.oetiker.ch/pipermail/rrd-developers/attachments/20090314/81798e9f/attachment.bin