| Message ID | 20191204113419.2298-1-sjpark@amazon.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | xen/blkback: Aggressively shrink page pools if a memory pressure is detected | expand |
> -----Original Message----- > From: Xen-devel <xen-devel-bounces@lists.xenproject.org> On Behalf Of > SeongJae Park > Sent: 04 December 2019 11:34 > To: konrad.wilk@oracle.com; roger.pau@citrix.com; axboe@kernel.dk > Cc: sj38.park@gmail.com; xen-devel@lists.xenproject.org; linux- > block@vger.kernel.org; linux-kernel@vger.kernel.org; Park, Seongjae > <sjpark@amazon.com> > Subject: [Xen-devel] [PATCH 0/2] xen/blkback: Aggressively shrink page > pools if a memory pressure is detected > > Each `blkif` has a free pages pool for the grant mapping. The size of > the pool starts from zero and be increased on demand while processing > the I/O requests. If current I/O requests handling is finished or 100 > milliseconds has passed since last I/O requests handling, it checks and > shrinks the pool to not exceed the size limit, `max_buffer_pages`. > > Therefore, `blkfront` running guests can cause a memory pressure in the > `blkback` running guest by attaching arbitrarily large number of block > devices and inducing I/O. OOI... How do guests unilaterally cause the attachment of arbitrary numbers of PV devices? Paul
On 04.12.19 12:52, Durrant, Paul wrote: >> -----Original Message----- >> From: Xen-devel <xen-devel-bounces@lists.xenproject.org> On Behalf Of >> SeongJae Park >> Sent: 04 December 2019 11:34 >> To: konrad.wilk@oracle.com; roger.pau@citrix.com; axboe@kernel.dk >> Cc: sj38.park@gmail.com; xen-devel@lists.xenproject.org; linux- >> block@vger.kernel.org; linux-kernel@vger.kernel.org; Park, Seongjae >> <sjpark@amazon.com> >> Subject: [Xen-devel] [PATCH 0/2] xen/blkback: Aggressively shrink page >> pools if a memory pressure is detected >> >> Each `blkif` has a free pages pool for the grant mapping. The size of >> the pool starts from zero and be increased on demand while processing >> the I/O requests. If current I/O requests handling is finished or 100 >> milliseconds has passed since last I/O requests handling, it checks and >> shrinks the pool to not exceed the size limit, `max_buffer_pages`. >> >> Therefore, `blkfront` running guests can cause a memory pressure in the >> `blkback` running guest by attaching arbitrarily large number of block >> devices and inducing I/O. > OOI... How do guests unilaterally cause the attachment of arbitrary numbers of PV devices? Good point. Many systems have their limit for the maximum number of the devices. Thus, 'arbitrarily' large number of devices cannot be attached. So, there is the upperbound. System administrators might be able to avoid the memory pressure problem by setting the limit low enough or giving more memory to the 'blkback' running guest. However, many systems also tempt to set the limit high enough so that guests can satisfy and to give minimal memory to the 'blkback' running guest for cost efficiency. I believe this patchset can be helpful for such situations. Anyway, using the term 'arbitrarily' is obvisously my fault. I will update the description in the next version of patchset. Thanks, SeongJae Park > > Paul >
Each `blkif` has a free pages pool for the grant mapping. The size of the pool starts from zero and be increased on demand while processing the I/O requests. If current I/O requests handling is finished or 100 milliseconds has passed since last I/O requests handling, it checks and shrinks the pool to not exceed the size limit, `max_buffer_pages`. Therefore, `blkfront` running guests can cause a memory pressure in the `blkback` running guest by attaching arbitrarily large number of block devices and inducing I/O. This patchset avoids such problematic situations by shrinking the pools aggressively (further the limit) for a while if a memory pressure is detected. Discussions =========== The shrinking mechanism returns only pages in the pool which are not currently be used by blkback. In other words, the pages that will be shrunk are not mapped with foreign pages. Because this patchset is changing only the shrink limit but uses the shrinking mechanism as is, this patchset does not introduce security issues such as improper unmappings. The first patch keeps the aggressive shrinking limit for one milisecond from last memory pressure detected time. The duration should be neither too short nor too long. If it is too long, free pages pool shrinking overhead can reduce the I/O performance. If it is too short, blkback will not free enough pages to reduce the memory pressure. I set the value as 1 millisecond by default because I believe that 1 millisecond is a short duration in terms of I/O while it is a long duration in terms of memory operations. Also, as the original shrinking mechanism works for every 100 milliseconds, this could be a somewhat reasonable choice. In actual, the default value worked well for my test (refer to below section for the detail of the test). Nevertheless, the proper duration would depends on given configurations and workloads. The second patch therefore allows users to set it via a module parameter interface. Memory Pressure Test ==================== To show whether this patchset fixes the above mentioned memory pressure situation well, I configured a test environment. On the `blkfront` running guest instances of a virtualized environment, I attach arbitrarily large number of network-backed volume devices and induce I/O to those. Meanwhile, I measure the number of pages that swapped in and out on the `blkback` running guest. The test ran twice, once for the `blkback` before this patchset and once for that after this patchset. Roughly speaking, this patchset has reduced those numbers 130x (pswpin) and 34x (pswpout) as below: pswpin pswpout before 76,672 185,799 after 587 5,402 Performance Overhead Test ========================= This patchset could incur I/O performance degradation under memory pressure because the aggressive shrinking will require more page allocations. To show the overhead, I artificially made an aggressive pages pool shrinking situation and measured the I/O performance of a `blkfront` running guest. For the artificial shrinking, I set the `blkback.max_buffer_pages` using the `/sys/module/xen_blkback/parameters/max_buffer_pages` file. We set the value to `1024` and `0`. The `1024` is the default value. Setting the value as `0` incurs the worst-case aggressive shrinking stress. For the I/O performance measurement, I use a simple `dd` command. Default Performance ------------------- [dom0]# echo 1024 > /sys/module/xen_blkback/parameters/max_buffer_pages [instance]$ for i in {1..5}; do dd if=/dev/zero of=file bs=4k count=$((256*512)); sync; done 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 11.7257 s, 45.8 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8827 s, 38.7 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8781 s, 38.7 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8737 s, 38.7 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8702 s, 38.7 MB/s Worst-case Performance ---------------------- [dom0]# echo 0 > /sys/module/xen_blkback/parameters/max_buffer_pages [instance]$ for i in {1..5}; do dd if=/dev/zero of=file bs=4k count=$((256*512)); sync; done 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 11.7257 s, 45.8 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.878 s, 38.7 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8746 s, 38.7 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8786 s, 38.7 MB/s 131072+0 records in 131072+0 records out 536870912 bytes (537 MB) copied, 13.8749 s, 38.7 MB/s In short, even worst case aggressive pools shrinking makes no visible performance degradation. I think this is due to the slow speed of the I/O. In other words, the additional page allocation overhead is hidden under the much slower I/O time. SeongJae Park (2): xen/blkback: Aggressively shrink page pools if a memory pressure is detected blkback: Add a module parameter for aggressive pool shrinking duration drivers/block/xen-blkback/blkback.c | 35 +++++++++++++++++++++++++++-- 1 file changed, 33 insertions(+), 2 deletions(-)