Re: [PATCH 00/11] Add compression support to pstore

[Aruna Balakrishnaiah <aruna@linux.vnet.ibm.com>]

Hi Tony,

On Wednesday 07 August 2013 08:55 AM, Tony Luck wrote:
> On Tue, Aug 6, 2013 at 6:58 PM, Aruna Balakrishnaiah
> <aruna@linux.vnet.ibm.com> wrote:
>> The patch looks right. I will clean it up. Does the issue still persist
>> after this?
> Things seem to be working - but testing has hardly been extensive (just
> a couple of forced panics).
>
> I do have one other question. In this code:
>
>>>                   if (compressed && (type == PSTORE_TYPE_DMESG)) {
>>>                           big_buf_sz = (psinfo->bufsize * 100) / 45;
> Where does the magic multiply by 1.45 come from?  Is that always enough
> for the decompression of "dmesg" type data to succeed?

I had this in my cover letter of the series, posting the same from it

Writing to persistent store
----------------------------
Compression will reduce the size of oops/panic report to atmost 45% of its
original size. (Based on experiments done while providing compression support
to nvram by Jim keniston).
Hence buffer of size ( (100/45 approx 2.22) *<registered_buffer> is allocated).
The compression parameters selected based on some experiments:
compression_level = 6, window_bits = 12, memory_level = 4  which achieved a
significant compression of 12 % of uncompressed buffer size tried upto 36k.
Data is compressed from the bigger buffer to registered buffer which is
returned to backends.
Pstore will indicate that with a flag 'compressed' which is passed to backends.
Using this flag, backends will add a flag in their header to indicate the data
is compressed or not while writing to persistent store.


The significant compression that I have mentioned had repeated occurrences in the
text. When I tried with plain text I saw compression of around 45% with compression
parameters I have used.

If the record size is fixed across all the backends then it would be easy to come
up with a pre defined set of compression parameters as well as the buffer size of
compressed/decompressed data based on experiments. In power as of now, the maximum size
of the record is 4k. So compression support on power was provided with multiply (100/45)
considering the maximum record size to be 4k.

How is it with erst and efivars?

- Aruna

> -Tony
>