questions about swap (partition and file)
John Johnstone
jjohnstone-freebsdquestions at tridentusa.com
Wed Feb 12 19:27:26 UTC 2020
On 2/11/20 11:21 AM, tech-lists wrote:
> Hi,
>
> On Tue, Feb 11, 2020 at 04:55:25PM +0100, Polytropon wrote:
>> On Tue, 11 Feb 2020 15:50:09 +0000, tech-lists wrote:
>>> Which is it better to have - swapfile partition or swapfile?
>>>
>>> I have both, on a SSD. Thing is, I can write 10x faster to the
>>> file than I can
>>> to the partition. With a 512b block size, it can write at 17MB/s
>>> to the
>>> partition, and about 120-130MB/s to the swapfile, as reported by dd.
>>
>> Probably a partition is better because it saves the subsystems
>> accessing it to deal with the filesystem layer. A swap device
>> does not have a filesystem, and it's not under VFS control, so
>> no overhead of that kind.
>
> But given there's no overhead, then why on the face of it is it slower
> than a
> swapfile?
It is the memory management part of an OS that writes and reads memory
areas to swap space. It does so directly without using dd or going
through the OS file system handling. So whatever testing you are doing
to try to benchmark the performance of things cannot be compared to what
the OS does when it is using swap space.
Also file system caching can dramatically change the performance numbers
that will be reported when doing file I/O. Maybe you are just measuring
130MB/s of I/O writes generated by dd that are going to the write cache
before they are written to disk. If that's the case you're just
measuring the speed of I/Os hitting memory and not the disk. Powerfail
your system 0.5 seconds into your test and then check and see if you
actually got 65MB of data written to your disk.
If you are seeing 17MB/s of writes going directly to a partition there
will be no caching involved. Although that also is not exactly what the
OS does it will be a much closer comparison to the OS swap I/O.
-
John J.
More information about the freebsd-questions
mailing list