What data exist in zero-length file? In sparsed file 'holes' areas does
not exists on lower levels, this just 'imagination' of filesystem. The
areas with data are threating normally and would cached to MaxIQ only if
reed often.
In other words, MaxIQ works on block level. And any block level have no
idea about sparsed files or filesystem or any other swaps created on
block device. Read requests to empty area will not come to block level.
Read requests to filled area and write requests to any area will come as
they come for any other files.
Those two things just have no any connection. (like your ethernet card
does not care about ssh key length in any way).
But sparsed files in product have a great performance penalty - in my
tests any write operations to sparsed areas reduce performance about
twice. Most visible test is filesystem creation. And in long terms many
sparsed files on same storage means higher fragmentation (and more
random io operations, and performance loss).
So, disk space is cheap, filesystems are resizable, so no reason for
using sparse in production environment.
N.B. For SSD this can be other question, I'm talking about old pain HDD
(and even MaxIQ does not change it principally, it just an emulation of
big memory cache).
В Чтв, 08/07/2010 в 16:56 -0500, Andy Pace пишет:
> Right, but the data does exist. What I'm wondering is if it will treat the
> sparse image file as 1 file, or will it cache on a block basis.
>
> > Do you often read holes from sparse?
>
> I'm not sure how to answer that.
>
> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of George Shuklin
> Sent: Thursday, July 08, 2010 4:36 PM
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-users] Xen Sparse Image Files and MaxIQ SSD
>
> That's simple: caching only if data exists.
>
> All 'holes' in spare file does not exists. They are just opinion of
> filesystem, but on block level there is no sparse.
>
> One more: MaxIQ cache most reed data. Do you often read holes from sparse?
>
>
> В Чтв, 08/07/2010 в 16:08 -0500, Andy Pace пишет:
> > Hello list!
> >
> >
> >
> > I’m currently storing Xen sparse image files for VM’s on remote
> > attached storage devices, and looking to increase the amount of IO
> > available on it. I’ve ran across MaxIQ SSD read acceleration caching,
> > and was wondering if this is a smart move, given the way we use the
> > sparse image files.
> >
> >
> >
> > My fear is that the MaxIQ will cache entire image files, instead of
> > the most commonly accessed files *INSIDE* the VM. Can anyone confirm
> > or deny my theory?
> >
> >
> >
> > I’ve talked with Adaptec and this is what they told me:
> >
> >
> >
> > --------
> >
> > It's hard to say exactly. Keep in mind that a RAID card isn't aware of
> > files, only blocks. So depending on how the blocks are accessed, the
> > MaxIQ product may or may not help you. My first impression with 30GB
> > files is that it may not be very helpful.
> >
> > --------
> >
> >
> >
> > So, that being said, right now we are accessing the image files using
> > the ‘file’ mode:
> >
> >
> >
> > <disk type='file' device='disk'>
> >
> > <driver name='file'/>
> >
> > <source file='/distributed/vms/29609/disk.img'/>
> >
> > <target dev='xvda' bus='xen'/>
> >
> > </disk>
> >
> >
> >
> > Will the MaxIQ work in the way I want it to, or will I have to change
> > the way we access the images? Or will it end up trying to cache entire
> > images onto the SSD? Has anyone had experience with this type of
> > application?
> >
> >
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-users
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home