Re: [RFC 0/3]block: An IOPS based ioscheduler

[Shaohua Li <shaohua.li@intel.com>]

On Mon, 2012-01-09 at 09:16 +1100, Dave Chinner wrote:
> On Fri, Jan 06, 2012 at 01:12:29PM +0800, Shaohua Li wrote:
> > On Thu, 2012-01-05 at 14:50 +0800, Shaohua Li wrote:
> > > On Wed, 2012-01-04 at 18:19 +1100, Dave Chinner wrote:
> > > > On Wed, Jan 04, 2012 at 02:53:37PM +0800, Shaohua Li wrote:
> > > > > An IOPS based I/O scheduler
> > > > > 
> > > > > Flash based storage has some different characteristics against rotate disk.
> > > > > 1. no I/O seek.
> > > > > 2. read and write I/O cost usually is much different.
> > > > > 3. Time which a request takes depends on request size.
> > > > > 4. High throughput and IOPS, low latency.
> > > > > 
> > > > > CFQ iosched does well for rotate disk, for example fair dispatching, idle
> > > > > for sequential read. It also has optimization for flash based storage (for
> > > > > item 1 above), but overall it's not designed for flash based storage. It's
> > > > > a slice based algorithm. Since flash based storage request cost is very
> > > > > low, and drive has big queue_depth is quite popular now which makes
> > > > > dispatching cost even lower, CFQ's slice accounting (jiffy based)
> > > > > doesn't work well. CFQ doesn't consider above item 2 & 3.
> > > > > 
> > > > > FIOPS (Fair IOPS) ioscheduler is trying to fix the gaps. It's IOPS based, so
> > > > > only targets for drive without I/O seek. It's quite similar like CFQ, but
> > > > > the dispatch decision is made according to IOPS instead of slice.
> > > > > 
> > > > > The algorithm is simple. Drive has a service tree, and each task lives in
> > > > > the tree. The key into the tree is called vios (virtual I/O). Every request
> > > > > has vios, which is calculated according to its ioprio, request size and so
> > > > > on. Task's vios is the sum of vios of all requests it dispatches. FIOPS
> > > > > always selects task with minimum vios in the service tree and let the task
> > > > > dispatch request. The dispatched request's vios is then added to the task's
> > > > > vios and the task is repositioned in the sevice tree.
> > > > > 
> > > > > The series are orgnized as:
> > > > > Patch 1: separate CFQ's io context management code. FIOPS will use it too.
> > > > > Patch 2: The core FIOPS.
> > > > > Patch 3: request read/write vios scale. This demontrates how the vios scale.
> > > > > 
> > > > > To make the code simple for easy view, some scale code isn't included here,
> > > > > some not implementated yet.
> > > > > 
> > > > > TODO:
> > > > > 1. ioprio support (have patch already)
> > > > > 2. request size vios scale
> > > > > 3. cgroup support
> > > > > 4. tracing support
> > > > > 5. automatically select default iosched according to QUEUE_FLAG_NONROT.
> > > > > 
> > > > > Comments and suggestions are welcome!
> > > > 
> > > > Benchmark results?
> > > I didn't have data yet. The patches are still in earlier stage, I want
> > > to focus on the basic idea first.
> > since you asked, I tested in a 4 socket machine with 12 X25M SSD jbod,
> > fs is ext4.
> > 
> > workload		percentage change with fiops against cfq
> > fio_sync_read_4k        -2
> > fio_mediaplay_64k       0
> > fio_mediaplay_128k      0
> > fio_mediaplay_rr_64k    0
> > fio_sync_read_rr_4k     0
> > fio_sync_write_128k     0
> > fio_sync_write_64k      -1
> > fio_sync_write_4k       -2
> > fio_sync_write_64k_create       0
> > fio_sync_write_rr_64k_create    0
> > fio_sync_write_128k_create      0
> > fio_aio_randread_4k     -4
> > fio_aio_randread_64k    0
> > fio_aio_randwrite_4k    1
> > fio_aio_randwrite_64k   0
> > fio_aio_randrw_4k       -1
> > fio_aio_randrw_64k      0
> > fio_tpch        9
> > fio_tpcc        0
> > fio_mmap_randread_4k    -1
> > fio_mmap_randread_64k   1
> > fio_mmap_randread_1k    -8
> > fio_mmap_randwrite_4k   35
> > fio_mmap_randwrite_64k  22
> > fio_mmap_randwrite_1k   28
> > fio_mmap_randwrite_4k_halfbusy  24
> > fio_mmap_randrw_4k      23
> > fio_mmap_randrw_64k     4
> > fio_mmap_randrw_1k      22
> > fio_mmap_randrw_4k_halfbusy     35
> > fio_mmap_sync_read_4k   0
> > fio_mmap_sync_read_64k  -1
> > fio_mmap_sync_read_128k         -1
> > fio_mmap_sync_read_rr_64k       5
> > fio_mmap_sync_read_rr_4k        3
> > 
> > The fio_mmap_randread_1k has regression against 3.2-rc7, but no
> > regression against 3.2-rc6 kernel, still checking why. The fiops has
> > improvement for read/write mixed workload. CFQ is known not good for
> > read/write mixed workload.
> 
> Numbers like this are meaningless without knowing what the hardware
> capability is and how the numbers compare to that raw capability.
> They tell me only mmap based random write improves in
> performance, and only one specific type of random write improves,
> not all types.
> 
> That raises more questions that it answers: why do AIO based random
> writes not go any faster? Is that because even with CFQ, AIO based
> random writes saturate the device?  i.e. is AIO based IO that much
> faster than mmap based IO that there is no scope for improvement on
> your hardware?
> 
> You need to present raw numbers and give us some idea of how close
> those numbers are to raw hardware capability for us to have any idea
> what improvements these numbers actually demonstrate.
Yes, your guess is right. The hardware has limitation. 12 SSD exceeds
the jbod capability, for both throughput and IOPS, that's why only
read/write mixed workload impacts. I'll use less SSD in later tests,
which will demonstrate the performance better. I'll report both raw
numbers and fiops/cfq numbers later.

Thanks,
Shaohua