Comments on: Do you need asmlib?

By: Riyaj Shamsudeen

Riyaj Shamsudeen — Wed, 13 Mar 2013 15:43:00 +0000

Hello Mladen
Thanks for reading my blog. Is there a documentation suggesting the increase in interrupts? Would you mind sharing more details about this increase in interrupts?
Even if udev/devmapper setup doubles interrupts, I doubt that interrupts are causing noticeable, even measurable, performance issues with the use of udev/device mapper/SCSI setup. I would be glad to see any valid performance benchmark comparing asmlib and udev/device mapper setup. I heard stories of few vendors performing benchmarks comparing these two setup, but no valid benchmark results were ever released.
Professionally, I have setup many high end Linux clusters using udev/device mapper( without ASMLIB) supporting enormous amount of workload and yet to see an issue due to udev/device mapper setup. Further, device mapper setup provides a) clean role separation as DBAs typically don’t need to be involved in disk setup b) no worries about Linux vendor support c) no worries about upgrading asmlib libraries during Linux upgrade.
So, No, I don’t consider interrupts as a big enough not to consider device mapper, due to huge operational convenience that this provide. I guess, we will be agreeing to disagree
Cheers
Riyaj

By: Mladen Gogala

Mladen Gogala — Mon, 11 Mar 2013 22:08:44 +0000

Hi Riyaj!
Nice article, however I would beg to differ. SCSI interface to the SATA disks, which is what UDEV does, and consequent using of Jorg Schiller’s libsg3 interface to do I/O on SCSI devices was invented in order to support today largely forgotten CD burners. The problem is, however, that this doubles the amount of interrupts needed to perform IO and therefore slows things down. I would recommend ASMLib, as it uses raw interface and doesn’t require additional interrupts.

By: Lucas Almeida

Lucas Almeida — Fri, 15 Feb 2013 15:38:48 +0000

Thank you Riyaj… very good!!!

By: Sergey Golikov

Sergey Golikov — Thu, 01 Nov 2012 10:05:27 +0000

Hi
What option i have to use in RHEL 6 ? Does the /etc/multipath.conf still valid on RHEL 6?

By: Jakub Wartak

Jakub Wartak — Tue, 04 Sep 2012 16:59:25 +0000

Hi Riyaj, please take a look on https://blogs.oracle.com/wim/entry/asmlib. Wim mentions that (if i understood properly) there is ongoing activity to have DIF/DIX/T10 storage validation done partially by ASMLib. I think it is something close to the old H.A.R.D. initiative (end-to-end data validation from RDBMS down to single drive). If that is true, then I think thing is going to be the killer feature of ASMLib+UEK combo … There is additional (older) presentation here https://oss.oracle.com/~mkp/docs/lpc08-data-integrity.pdf . The sad part about ASMLib is that it appears that Oracle officially is recommending shutting down whole CRS stack & unloading ASMLib on all nodes when you are doing kernel upgrades…

By: nilesh nayak

nilesh nayak — Thu, 30 Aug 2012 03:55:09 +0000

Very Nice

By: Vyacheslav Rasskazov

Vyacheslav Rasskazov — Wed, 29 Aug 2012 22:47:40 +0000

Thank you for useful article.
I think, needs to say that uid,gid,mode attributes in /etc/multipath.conf no longer works at RHEL/OEL 6.
Device persistence is not mandatory condition for ASM, because ASM reads disk headers and asm_diskstring like /dev/mapper will be ok. With multipath configuration device persistence also can be achieved by setting user_friendly_names settings to “no”.