Re: [zfs-discuss] ZFS dump and swap

2008-09-25 Thread Richard Elling
Kyle McDonald wrote:
 Darren J Moffat wrote:
   
 John Cecere wrote:
   
 
 The man page for dumpadm says this:

 A given ZFS volume cannot be configured for both the swap area and the dump 
 device.

 And indeed when I try to use a zvol as both, I get:

 zvol cannot be used as a swap device and a dump device

 My question is, why not ?
 
   
 Swap is a normal ZVOL and subject to COW, checksum, compression (and 
 coming soon encryption).

   
 
 Would there be no performance benefits from having swap read/write from 
 contiguous preallocated space also?
   

Not really, if you have to swap, you have no performance. Period.
You lose about 3 orders of magnitude of memory latency by going to
disk.  If one is 2x faster than another, you still can't regain the orders
of magnitude.

 I do realize that nifty features like encryption might be lost in that 
 case, but Im wondering if there's any performance to be gained?

 Then again if you're concerned about performance you need to just buy 
 ram till you stop swapping all together, huh?
   

Yes.  Fast, inexpensive, reliable: pick two.
 -- richard

___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] ZFS dump and swap

2008-09-25 Thread Darren J Moffat
Richard Elling wrote:
 Would there be no performance benefits from having swap read/write 
 from contiguous preallocated space also?
   
 
 Not really, if you have to swap, you have no performance. Period.
 You lose about 3 orders of magnitude of memory latency by going to
 disk.  If one is 2x faster than another, you still can't regain the orders
 of magnitude.

In addition to that you REALLY REALLY want protection of what ever you 
write to a swap device, so copies=2 or mirror/raidz pool config.

Prior to ZFS the recommended best practice was to mirror your swap with 
a volume manager such as SVM.

Having swap compressed could insome cases reduce the impact because you 
will do less IO to get more pages of RAM pagedout.

Plus you really do want your swap encrypted if you have any encrypted 
datasets - otherwise you may have defeated the point of encrypted the 
datasets (depends on your whole system config and if your OS pool and 
data pool are the same and have the same physical access risks).

--
Darren J Moffat
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] ZFS dump and swap

2008-09-24 Thread Darren J Moffat
John Cecere wrote:
 The man page for dumpadm says this:
 
 A given ZFS volume cannot be configured for both the swap area and the dump 
 device.
 
 And indeed when I try to use a zvol as both, I get:
 
 zvol cannot be used as a swap device and a dump device
 
 My question is, why not ?

Swap is a normal ZVOL and subject to COW, checksum, compression (and 
coming soon encryption).

Dump ZVOLs are preallocated contiguous space that are written to 
directly by the ldi_dump routines, they aren't written to by normal ZIO 
transactions, they aren't checksum'd - the compression is done by the 
dump layer not by ZFS.  This is needed because when we are writing a 
crash dump we want as little as possible in IO the stack.

--
Darren J Moffat
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] ZFS dump and swap

2008-09-24 Thread John Cecere
Darren,

Thanks for the explanation. Would you object if I opened a bug on the zfs man 
page to include what you've written here ?

Thanks again,
John


Darren J Moffat wrote:
 John Cecere wrote:
 The man page for dumpadm says this:

 A given ZFS volume cannot be configured for both the swap area and the 
 dump device.

 And indeed when I try to use a zvol as both, I get:

 zvol cannot be used as a swap device and a dump device

 My question is, why not ?
 
 Swap is a normal ZVOL and subject to COW, checksum, compression (and 
 coming soon encryption).
 
 Dump ZVOLs are preallocated contiguous space that are written to 
 directly by the ldi_dump routines, they aren't written to by normal ZIO 
 transactions, they aren't checksum'd - the compression is done by the 
 dump layer not by ZFS.  This is needed because when we are writing a 
 crash dump we want as little as possible in IO the stack.
 
 -- 
 Darren J Moffat

-- 
John Cecere
Americas Technology Office / Sun Microsystems
732-302-3922 / [EMAIL PROTECTED]
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] ZFS dump and swap

2008-09-24 Thread Darren J Moffat
John Cecere wrote:
 Darren,
 
 Thanks for the explanation. Would you object if I opened a bug on the 
 zfs man page to include what you've written here ?

I don't know if what I said is considered implementation detail or not. 
  Feel free to log the bug but I can't say either way if it would be 
considered appropriate for the man page.

--
Darren J Moffat
___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss


Re: [zfs-discuss] ZFS dump and swap

2008-09-24 Thread Kyle McDonald
Darren J Moffat wrote:
 John Cecere wrote:
   
 The man page for dumpadm says this:

 A given ZFS volume cannot be configured for both the swap area and the dump 
 device.

 And indeed when I try to use a zvol as both, I get:

 zvol cannot be used as a swap device and a dump device

 My question is, why not ?
 

 Swap is a normal ZVOL and subject to COW, checksum, compression (and 
 coming soon encryption).

   
Would there be no performance benefits from having swap read/write from 
contiguous preallocated space also?

I do realize that nifty features like encryption might be lost in that 
case, but Im wondering if there's any performance to be gained?

Then again if you're concerned about performance you need to just buy 
ram till you stop swapping all together, huh?

   -Kyle

 Dump ZVOLs are preallocated contiguous space that are written to 
 directly by the ldi_dump routines, they aren't written to by normal ZIO 
 transactions, they aren't checksum'd - the compression is done by the 
 dump layer not by ZFS.  This is needed because when we are writing a 
 crash dump we want as little as possible in IO the stack.

 --
 Darren J Moffat
 ___
 zfs-discuss mailing list
 zfs-discuss@opensolaris.org
 http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
   

___
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss