Re: [PATCH 3/3] zram: rework reported to end-user zram statistics

[Jerome Marchand]

On 01/14/2014 12:10 PM, Sergey Senozhatsky wrote:
> On (01/14/14 11:38), Jerome Marchand wrote:
>> On 01/14/2014 10:37 AM, Sergey Senozhatsky wrote:
>>> 1) Introduce ZRAM_ATTR_RO macro to generate zram atomic64_t stats
>>> `show' functions and reduce code duplication.
>>>
>>> 2) Account and report back to user numbers of failed READ and WRITE
>>> operations.
>>>
>>> 3) Remove `good' and `bad' compressed sub-requests stats. RW request may
>>> cause a number of RW sub-requests. zram used to account `good' compressed
>>> sub-queries (with compressed size less than 50% of original size), `bad'
>>> compressed sub-queries (with compressed size greater that 75% of original
>>> size), leaving sub-requests with compression size between 50% and 75% of
>>> original size not accounted and not reported.
>>
>> That's weird: good/bad_compress are accounted, but it seems to me that
>> they are to never used in any way. If so, there is indeed no reason to
>> keep them.
>>
>>
>>> Account each sub-request
>>> compression size so we can calculate real device compression ratio.
>>
>> Your patch doesn't change the way pages_stored and compr[essed]_size
>> are accounted. What does your patch change that allow us to calculate
>> the "real" compression ratio?
>>
>>>
>>> 4) reported zram stats:
>>>   - num_writes  -- number of writes
>>>   - num_reads  -- number of reads
>>>   - pages_stored -- number of pages currently stored
>>>   - compressed_size -- compressed size of pages stored
>>
>> Wouldn't it be more practical to report the original and compressed
>> data sizes using the same units as it is currently done?
>>
> 
> hm, do we really need pages_stored stats? what kind of unseful information it
> shows to end user?.. perhaps, it's better to replace it with accounted passed
> bvec->bv_len (as uncompressed_size).
> 

That's really going to complicates things. We would need to keep track
of which sectors of a particular page has been written to. It's much
easier to keep current page granularity and consider any partial I/O
as an whole page I/O.

> 	-ss
> 
>> Jerome
>>
>>>   - pages_zero  -- number of zero filled pages
>>>   - failed_read -- number of failed reads
>>>   - failed_writes -- can happen when memory is too low
>>>   - invalid_io   -- non-page-aligned I/O requests
>>>   - notify_free  -- number of swap slot free notifications
>>>   - memory_used -- zs pool zs_get_total_size_bytes()
>>>
>>> Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@xxxxxxxxx>


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/