Re: [PATCH] mm: add zblock allocator

[Shakeel Butt <shakeel.butt@linux.dev>]

Hi Igor,

On Wed, Apr 02, 2025 at 10:45:25AM +0400, igor.b@beldev.am wrote:
> Hi Shakeel,
> 
> 2025-04-02 03:16 Shakeel Butt wrote:
> > Hi Vitaly,
> > 
> > On Tue, Apr 01, 2025 at 07:17:54PM +0200, Vitaly Wool wrote:
> > > zblock is a special purpose allocator for storing compressed pages.
> > > It stores integer number of compressed objects per its block. These
> > > blocks consist of several physical pages (2**n, i. e. 1/2/4/8).
> > > 
> > > With zblock, it is possible to densely arrange objects of various
> > > sizes
> > > resulting in low internal fragmentation. Also this allocator tries to
> > > fill incomplete blocks instead of adding new ones,  in many cases
> > > providing a compression ratio substantially higher than z3fold and
> > > zbud
> > > (though lower than zmalloc's).
> > > 
> > > zblock does not require MMU
> > 
> > Can you explain why not requiring MMU is important for your use-case?
> > Also what exactly is your use-case? Are you planning to use zblock
> > through zram or zswap or something new?
> 
> We have 2 use cases for a zpool backend: data centers (zswap) and small MMU
> less devices, this is where zram comes ito play. We know that zram over
> zpool patch is still out of tree but we also know the story behind that :)
>

So zswap and zram are the use-cases. It is usually frowned upon to add a
general purpose library without an actual user. So, I would recommend to
add a user along with the zblock.

> > > to operate and also is superior to zsmalloc
> > > with regard to average performance and worst execution times, thus
> > > allowing for better response time and real-time characteristics of the
> > > whole system.
> > > 
> > > E. g. on a series of stress-ng tests run on a Raspberry Pi 5, we get
> > > 5-10% higher value for bogo ops/s in zblock/zsmalloc comparison.
> > 
> > Can you explain a bit more on this test? How is this test using zblock?
> 
> We're running stress-ng on a number of devices. The results published in the
> previous message are from the run on Raspberry Pi 5. zpool's settings used
> for comparison are lz4/80%.
> 

My question was mainly if this is zswap or zram and I think it is zswap
based on the stress-ng command. For zswap, I would recommend to run perf
experiment on a server (or beefy desktop) as well. Also pick a normal
server like workload. I would be interested in looking at zswapin
latencies and zswapout throuput in addition to the metrics the workload
cares about.

> Thanks,
> Igor
>