schrieb Harald Schmalzbauer am 10.07.2011 16:28 (localtime):
> schrieb Adrian Reyer am 10.07.2011 15:58 (localtime):
> ...
>> I found using Spool Data for copy jobs to be faster for my setup. I have
>> fast local disks for spooling, but some of my disk storage is accessed
>> vie iSCSI on 1-GBit/s-links.
>> However, I am currently running a few copy jobs and the limiting factor
>> seems to be my bacula-sd consuming 1 complete CPU, throttling me at
>> 55MB/s. The CPU is an older 'AMD Athlon(tm) 64 X2 Dual Core Processor
>> 3800+'
>
> Hmmm, interesting point. Is the storage daemon single threaded? I'll
> check this, since I have an "old" dual core Xeon, but I checked that
> there were at least 25% idle of CPU time (ZFS compression is used, so
> that client-to-disk backup is uncompressed avoiding interference with
> later (migration) tape drive compression)
>
>>> Maybe this was not an issue with slower tape drives. LTO2 would only
>>> suffer from about 6% performance loss, if my wild guess has any truth...
>>
>> LTO4 as well here, and no ear next to the drive. However, 'mt status'
>> won't run as the drive is in use by the copy jobs, how you got that
>> info?
>
> On FreeBSD there's a special control device for every sequential access
> device (/dev/sa0 /dev/sa0.ctl for example).
>
> No luck so far finding technical end-user-details for the design of LTO
> drives. I'm really wondering if file marks are written to the tape or
> just in the cartridge memory chip. And even if they get written to the
> tape, is it unavoidable that streaming is interrupted???
But reading the great official bacula manual answered one of my
questions and prooved my wild guess to be correct.
in Chapter 19, page 188, one can find:
Maximum File Size = size No more than size bytes will be written into a
given logical file on the volume.
Once this size is reached, an end of file mark is written on the
volume and subsequent data are written into the next file.
...
If you are configuring an LTO-3 or LTO-4 tape, you probably will
want to set the Maximum File Size to 2GB to avoid making the drive
stop to write an EOF mark.
...
I set it to 100G and the frequent interruptions vanished :-)
Maybe the suggested 2GB are well chosen for LTO-3, I think for LTO-4 you
need much larger values than 2GB.
Now I still have the problem that I don't get more than 110MB/s using
bacula to the drive, while %busy states 55% and the disks from the ZFS
pool only read 15MB/s, reflecting that currently written material is
compresed by almost 2:1. I have seen transerrates of 150MB/s with tar...
CPU is 83% idle, ZFS disks <10% busy and the tape drive ~50% idle. Why
don't I get the 150MB/s as seen with tar ...?!?...
I'll report if I find out and thank you in advance for any hints.
-Harry
signature.asc
Description: OpenPGP digital signature
------------------------------------------------------------------------------
All of the data generated in your IT infrastructure is seriously valuable.
Why? It contains a definitive record of application performance, security
threats, fraudulent activity, and more. Splunk takes this data and makes
sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-d2d-c2 _______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users
|
