In the message dated: Thu, 01 Dec 2011 16:27:33 GMT,
The pithy ruminations from Alan Brown on
<Re: [Bacula-users] tuning lto-4> were:
=> gary artim wrote:
=> > You guys/gals are great, very responsive! I did try
=> > spooling/despooling and my run times shot up.
=>
=> They will - you're copying everything twice (disk to disk to tape), but
=> this is the only way to achieve fast despooling speeds - if you don't do
=> this then your LTO drive will start to "shoe shine" and speeds drop off
=> rapidly when it happens.
And you increase wear & tear on the drive and media.
=>
=> The trick is to run multiple jobs at once - you have to spool to achieve
=> this anyway or extracting will be a nightmare.
=>
=> Spooling is a net gain when you're running incrementals.
=>
Not necessarily. Spooling is a gain if you are measuring the speed
of writing to tape. Spooling may be a net loss for end-to-end (client
machine-->spool server-->tape drive) speed.
For backups clients where the total volume being backed up is less than
the spool size, then there's a very good chance of a performance gain. As
soon as a job requires multiple rounds of spooling and de-spooling,
there's a good chance of a performance loss because bacula stops reading
from the client machine (stops spooling that job) as soon as despooling
begins. Of course, spooling allows you to run multiple jobs in parallel, a
clear win over running them in series.
See:
[1] http://copilotco.com/mail-archives/bacula-devel.2007/msg02642.html
[2]
http://www.bacula.org/git/cgit.cgi/bacula/plain/bacula/projects?h=Branch-5.1
[3]
http://www.mail-archive.com/bacula-users@lists.sourceforge.net/msg49366.html
=> Spooling MUST happen on a fast dedicated drive. You're best off dropping
=> in a fast SSD such as a 64/128Gb OCZ vertex3 or similar to handle it.
Hmm...for LTO4 (large spool files are good), you might want more space
than that, particularly if you have multiple clients (multiple spool
files). A more cost-effective option might be several fast drives (10K
or 15K SAS or SCSI) in RAID-0. It doesn't take very many drives in RAID0 to
have an aggregate drive throughput that is greater than the bus interface.
=>
=> > I was using a simple
=> > 7200 drive though, no ssd or raid...I assume the performance gain
Yeah, the sustained read speed from a 7.2k RPM drive is lower than the
possible write speed to an LTO-4 drive:
http://www.seagate.com/www/en-us/support/before_you_buy/speed_considerations
=> > happens when your networks multi machines...wearing multiple hats so
=> > will report back on btape next week, unless I get some time. gary
=>
=> Even on a single host, if the heads are thrashing then spooling will
=> save time overall. The big advantage is being able to run multiple jobs
=> so that several are spooling data at the same time one is despooling.
Absolutely. Spooling is a big win for multiple jobs, and for reducing
wear&tear on the tape drive. It may or may not give a performance increase for
any single backup job.
Mark
------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure
contains a definitive record of customers, 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-novd2d
_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users
