Hi, I have installed Xen on Linux 2.6.11.10 <http://2.6.11.10> and i am trying to do Asynchronous Direct IO on SAS drives. The application which does the asynchronous direct io on SAS drive is running on Domain 0. Actually the IOPs what i get for a 512Bytes IO size is 67, but if i do the same operation on Linux 2.6.11.10 <http://2.6.11.10> native kernel, i get 267 IOPs.Cananyone tell me why this huge differnece? Am i missing something? In the current setup on Xen, if i do Synchronous IO, then i am getting 265 IOPs which is expected. So i am wondering why Asynchronous IO should behave this way? Is there any reason?? Thanks, Priya _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> I have installed Xen on Linux 2.6.11.10 and i am trying to do > Asynchronous Direct IO on SAS drives. The application which > does the asynchronous direct io on SAS drive is running on > Domain 0. Actually the IOPs what i get for a 512Bytes IO size > is 67, but if i do the same operation on Linux 2.6.11.10 > native kernel, i get 267 IOPs.Can anyone tell me why this > huge differnece? Am i missing something? In the current setup > on Xen, if i do Synchronous IO, then i am getting 265 IOPs > which is expected. So i am wondering why Asynchronous IO > should behave this way? Is there any reason??That''s odd. You might want to try the -unstable tree. I know Andy has used AIO just fine on -unstable. Ian _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi, I tried the same operation with unstable version too. I have changed the IO scheduler to atropos and tried. But no use. I always get the same results. Has anyone checked the Asynchronous IO path using libaio? It would be very much helpful if you can give me some ideas to proceed further, Thanks, Priya. On 9/8/05, Ian Pratt <m+Ian.Pratt@cl.cam.ac.uk> wrote:> > > I have installed Xen on Linux 2.6.11.10 <http://2.6.11.10> and i am > trying to do > > Asynchronous Direct IO on SAS drives. The application which > > does the asynchronous direct io on SAS drive is running on > > Domain 0. Actually the IOPs what i get for a 512Bytes IO size > > is 67, but if i do the same operation on Linux 2.6.11.10<http://2.6.11.10> > > native kernel, i get 267 IOPs.Can anyone tell me why this > > huge differnece? Am i missing something? In the current setup > > on Xen, if i do Synchronous IO, then i am getting 265 IOPs > > which is expected. So i am wondering why Asynchronous IO > > should behave this way? Is there any reason?? > > That''s odd. You might want to try the -unstable tree. I know Andy has > used AIO just fine on -unstable. > > Ian >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi, Can anyone please explain to me how exactly the IO is handled if i have only Domain 0? Thanks, Priya. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> Can anyone please explain to me how exactly the IO is handled if i have > only Domain 0?Domain 0 runs the device drivers, which are permitted to write/read the devices in your system directly. All Xen really does is take the real physical interrupts and issue corresponding virtual interrupts to dom0 (which are then handled as usual). Xen also handles various PCI functions but they shouldn''t really matter once the system is up and running. HTH, Mark _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi Priya, I regularly use libaio in domain zero as a user-space backend for other domains and am able to saturate a MPT fusion at about 60MB/s without trying too hard. I seem to remember seeing a comment about a recent performance drop on the linux-aio list, possibly from 2.6.11 to 2.6.12, you might want to take a peek at that. Also, are you sure that your XenLinux dom0 kernel has your disk driver in it, and that it isn''t deferring to a less-efficient means of accessing the disk? a. On 9/9/05, Priya PM <pmpriya@gmail.com> wrote:> Hi, > > I tried the same operation with unstable version too. I have changed the IO > scheduler to atropos and tried. But no use. I always get the same results. > Has anyone checked the Asynchronous IO path using libaio? > > It would be very much helpful if you can give me some ideas to proceed > further, > > Thanks, > Priya. > > > > On 9/8/05, Ian Pratt <m+Ian.Pratt@cl.cam.ac.uk> wrote: > > > I have installed Xen on Linux 2.6.11.10 and i am trying to do > > > Asynchronous Direct IO on SAS drives. The application which > > > does the asynchronous direct io on SAS drive is running on > > > Domain 0. Actually the IOPs what i get for a 512Bytes IO size > > > is 67, but if i do the same operation on Linux 2.6.11.10 > > > native kernel, i get 267 IOPs.Can anyone tell me why this > > > huge differnece? Am i missing something? In the current setup > > > on Xen, if i do Synchronous IO, then i am getting 265 IOPs > > > which is expected. So i am wondering why Asynchronous IO > > > should behave this way? Is there any reason?? > > > > That''s odd. You might want to try the -unstable tree. I know Andy has > > used AIO just fine on -unstable. > > > > Ian > > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi, Thanks for your response. Even I got 52MB/sec when I tried to do IO with buffer size 1 MB. But if I try to do IO with buffer size 512 Bytes, I am getting 0.032 MB which is 67 IOPs which is not the expected result. Have you tried the asynchronosu IO with different IO sizes?? I am sure I am using right version of MPT driver in Domain 0. Moreover the same driver performs better if I do synchronous IO in Domain 0. I am confused. Thanks, Priya. -----Original Message----- From: xen-users-bounces@lists.xensource.com [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of Andrew Warfield Sent: Thursday, September 08, 2005 11:11 PM To: Priya PM Cc: Ian Pratt; ian.pratt@cl.cam.ac.uk; Xen-users@lists.xensource.com Subject: Re: [Xen-users] Asynchronous IO Hi Priya, I regularly use libaio in domain zero as a user-space backend for other domains and am able to saturate a MPT fusion at about 60MB/s without trying too hard. I seem to remember seeing a comment about a recent performance drop on the linux-aio list, possibly from 2.6.11 to 2.6.12, you might want to take a peek at that. Also, are you sure that your XenLinux dom0 kernel has your disk driver in it, and that it isn''t deferring to a less-efficient means of accessing the disk? a. On 9/9/05, Priya PM <pmpriya@gmail.com> wrote:> Hi, > > I tried the same operation with unstable version too. I have changed > the IO scheduler to atropos and tried. But no use. I always get thesame results.> Has anyone checked the Asynchronous IO path using libaio? > > It would be very much helpful if you can give me some ideas to proceed> further, > > Thanks, > Priya. > > > > On 9/8/05, Ian Pratt <m+Ian.Pratt@cl.cam.ac.uk> wrote: > > > I have installed Xen on Linux 2.6.11.10 and i am trying to do > > > Asynchronous Direct IO on SAS drives. The application which does > > > the asynchronous direct io on SAS drive is running on Domain 0. > > > Actually the IOPs what i get for a 512Bytes IO size is 67, but if > > > i do the same operation on Linux 2.6.11.10 native kernel, i get > > > 267 IOPs.Can anyone tell me why this huge differnece? Am i missing> > > something? In the current setup on Xen, if i do Synchronous IO, > > > then i am getting 265 IOPs which is expected. So i am wondering > > > why Asynchronous IO should behave this way? Is there any reason?? > > > > That''s odd. You might want to try the -unstable tree. I know Andy > > has used AIO just fine on -unstable. > > > > Ian > > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi Priya, I''m generally reading and writing 4KB buffers and haven''t done a lot of experimenting with other sizes. I you want to send the source that you are using to test throughput with, I''ll take a look at how it performs on my test box and see if I can help sort this out. cheers, a. On 9/9/05, Priya, PM <pm.priya@hp.com> wrote:> Hi, > > Thanks for your response. Even I got 52MB/sec when I tried to do IO with > buffer size 1 MB. But if I try to do IO with buffer size 512 Bytes, I am > getting 0.032 MB which is 67 IOPs which is not the expected result. Have > you tried the asynchronosu IO with different IO sizes?? > > I am sure I am using right version of MPT driver in Domain 0. Moreover > the same driver performs better if I do synchronous IO in Domain 0. I am > confused. > > Thanks, > Priya. > > -----Original Message----- > From: xen-users-bounces@lists.xensource.com > [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of Andrew > Warfield > Sent: Thursday, September 08, 2005 11:11 PM > To: Priya PM > Cc: Ian Pratt; ian.pratt@cl.cam.ac.uk; Xen-users@lists.xensource.com > Subject: Re: [Xen-users] Asynchronous IO > > Hi Priya, > > I regularly use libaio in domain zero as a user-space backend for > other domains and am able to saturate a MPT fusion at about 60MB/s > without trying too hard. I seem to remember seeing a comment about a > recent performance drop on the linux-aio list, possibly from 2.6.11 to > 2.6.12, you might want to take a peek at that. Also, are you sure that > your XenLinux dom0 kernel has your disk driver in it, and that it isn''t > deferring to a less-efficient means of accessing the disk? > > a. > > On 9/9/05, Priya PM <pmpriya@gmail.com> wrote: > > Hi, > > > > I tried the same operation with unstable version too. I have changed > > the IO scheduler to atropos and tried. But no use. I always get the > same results. > > Has anyone checked the Asynchronous IO path using libaio? > > > > It would be very much helpful if you can give me some ideas to proceed > > > further, > > > > Thanks, > > Priya. > > > > > > > > On 9/8/05, Ian Pratt <m+Ian.Pratt@cl.cam.ac.uk> wrote: > > > > I have installed Xen on Linux 2.6.11.10 and i am trying to do > > > > Asynchronous Direct IO on SAS drives. The application which does > > > > the asynchronous direct io on SAS drive is running on Domain 0. > > > > Actually the IOPs what i get for a 512Bytes IO size is 67, but if > > > > i do the same operation on Linux 2.6.11.10 native kernel, i get > > > > 267 IOPs.Can anyone tell me why this huge differnece? Am i missing > > > > > something? In the current setup on Xen, if i do Synchronous IO, > > > > then i am getting 265 IOPs which is expected. So i am wondering > > > > why Asynchronous IO should behave this way? Is there any reason?? > > > > > > That''s odd. You might want to try the -unstable tree. I know Andy > > > has used AIO just fine on -unstable. > > > > > > Ian > > > > > > > > > _______________________________________________ > > Xen-users mailing list > > Xen-users@lists.xensource.com > > http://lists.xensource.com/xen-users > > > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Reasonably Related Threads
- Gluster very poor performance when copying small files (1x (2+1) = 3, SSD)
- Gluster very poor performance when copying small files (1x (2+1) = 3, SSD)
- small files performance
- [PATCH] block: virtio_blk: don't hold spin lock during world switch
- [PATCH] block: virtio_blk: don't hold spin lock during world switch