UFS journal size
Quoting the manpage: -s jsize Specifies size of the journal if only one provider is used for both data and journal. The default is one gigabyte. Size should be chosen based on provider's load, and not on its size; recommended minimum is twice the size of the physical memory installed. It is not recommended to use gjournal for small file systems (e.g.: only few gigabytes big). My question is: if I have 4 or 8 GB of RAM should I create 8 or even 16 GB journals?.. This seems huge especially if the fs size without journal is only 10 gigs. Or the recommended minimum is for systems low on RAM? ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: UFS journal size
On 21/09/2011 10:48, Ross wrote: Quoting the manpage: -s jsize Specifies size of the journal if only one provider is used for both data and journal. The default is one gigabyte. Size should be chosen based on provider's load, and not on its size; recommended minimum is twice the size of the physical memory installed. It is not recommended to use gjournal for small file systems (e.g.: only few gigabytes big). My question is: if I have 4 or 8 GB of RAM should I create 8 or even 16 GB journals?.. This seems huge especially if the fs size without journal is only 10 gigs. Or the recommended minimum is for systems low on RAM? How much churn do you expect in the data on that partition? A journal that's about the same size as the actual filesystem in question and on the same physical device is not really going to get you any advantages. If it's mostly going to be read rather than written, then you wouldn't fill up that size of journal in any case. The 'twice physical RAM' advice is all about achieving maximum performance on large filesystems with lots of data writes: if write performance is not actually a limiting factor, then you could get away with a much smaller or even no journal at all. You might just as well use plain UFS+Softupdates. Softupdates to provide the meta-data ordering feature, so that if you do crash and need to fsck the filesystem, there's not going to be any really nasty stuff to fix. Plain UFS because a filesystem of that size will take about as long to fsck as it would to replay all the journalled but uncommitted updates. Cheers, Matthew -- Dr Matthew J Seaman MA, D.Phil. 7 Priory Courtyard Flat 3 PGP: http://www.infracaninophile.co.uk/pgpkey Ramsgate JID: matt...@infracaninophile.co.uk Kent, CT11 9PW signature.asc Description: OpenPGP digital signature
Re: UFS journal size
On Wed, 21 Sep 2011 11:41:08 +0100 Matthew Seaman wrote: On 21/09/2011 10:48, Ross wrote: My question is: if I have 4 or 8 GB of RAM should I create 8 or even 16 GB journals?.. This seems huge especially if the fs size without journal is only 10 gigs. Or the recommended minimum is for systems low on RAM? The 'twice physical RAM' advice is all about achieving maximum performance on large filesystems with lots of data writes: IIRC the original justification for 2*ram was as a crude rule-of-thumb to avoid panics. I think the idea was that writing the whole ram into one of the two journalling areas was an extreme case. You might just as well use plain UFS+Softupdates. Softupdates to provide the meta-data ordering feature, so that if you do crash and need to fsck the filesystem, there's not going to be any really nasty stuff to fix. And in 9.x UFS filesystems (even existing ones) will be able to use journalled soft-updates. This should give a fast fsck without the overheads of full data journalling or background fsck. Plain UFS because a filesystem of that size will take about as long to fsck as it would to replay all the journalled but uncommitted updates. FWIW fsck doesn't replay the journal, it just does a quick check for orphaned files and marks the filesystem as clean - uncommitted updates are left for gjournal. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: UFS journal size
On 21/09/2011 11:48, Ross wrote: Quoting the manpage: -s jsize Specifies size of the journal if only one provider is used for both data and journal. The default is one gigabyte. Size should be chosen based on provider's load, and not on its size; recommended minimum is twice the size of the physical memory installed. It is not recommended to use gjournal for small file systems (e.g.: only few gigabytes big). My question is: if I have 4 or 8 GB of RAM should I create 8 or even 16 GB journals?.. This seems huge especially if the fs size without journal is only 10 gigs. Or the recommended minimum is for systems low on RAM? You are probably missing that gjournal does full data journaling, not just metadata as is more common with other systems, so you need a journal which can hold *at least* everything you can hold in RAM, and then twice as much since one half may be replaying while other is being filled up. Even this is an estimate since it's heavily load-dependant. Or you can wait for 9.0 with metadata journaling. signature.asc Description: OpenPGP digital signature
Re: UFS journal size
On Wed, Sep 21, 2011 at 12:48:25PM +0300, Ross wrote: Quoting the manpage: -s jsize Specifies size of the journal if only one provider is used for both data and journal. The default is one gigabyte. Size should be chosen based on provider's load, and not on its size; recommended minimum is twice the size of the physical memory installed. It is not recommended to use gjournal for small file systems (e.g.: only few gigabytes big). My question is: if I have 4 or 8 GB of RAM should I create 8 or even 16 GB journals?.. This seems huge especially if the fs size without journal is only 10 gigs. Or the recommended minimum is for systems low on RAM? My experience has shown that speed of the underlying filessystem has a huge impact on the required size of the journal. I have a system running hardware RAID-10 on a 3Ware SATA controller. On a 100G partition, rsync would regularly cause a panic until I got my journal up above 10G. This particular host has only 1G of RAM and a single 3.4GHz P4 CPU. Sizing this particular box using gjournal was painful until I got the journal sizes large enough. It turns out the journals had to be so large (for the infrequent write burst) that a significant amount of disk was chewed up for journals that were mostly unused. If I had to do it over again, I would have not used gjournal and simply used softupdates. YMMV -- Regards, Doug ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
