public inbox for nncp-devel@lists.cypherpunks.su
Atom feed
From: John Goerzen <jgoerzen@complete•org>
To: Sergey Matveev <stargrave@stargrave•org>
Cc: nncp-devel@lists.cypherpunks.ru
Subject: Re: NNCP discussion on comp.mail.uucp
Date: Tue, 05 Jan 2021 10:01:58 -0600 [thread overview]
Message-ID: <874kjv74p5.fsf@complete.org> (raw)
In-Reply-To: <X/SAm1s5dInAlPcN@stargrave.org>
On Tue, Jan 05 2021, Sergey Matveev wrote:
>>NNCP automatic tossing.
>>also had automatic calling
>
> Both of these looks like easy to add.
Sweet!
> Here I just want to add trivial comments for
> https://floss.social/@jgoerzen/105499269925205743
>
> * Protection against Silent Data Corruption. Just want to
> mention that
> automatic self-healing (data correction) is performed also on
> a setup
> with just single disk. "Data" blocks, by default has just
> single copy
> and obviously they can not be repaired without redundancy. But
> metadata
> blocks ("inodes", dataset/zvol descriptions, block pointers,
> and so on)
> by default has copies=2 and physically they have two copies on
> the
> disk. Very important metadata blocks (meta object set (MOS),
> überblock) has copies=3. So metadata has (by default)
> redundancy and
> can be repaired by reading the copy from another part of the
> disk
Very true indeed.
> * Built-in #compression with fast algorithms. Compression (at
> least the
> most simple rle one) should be enabled also because sparse
> (zero-filled) blocks won't be allocated at all. If no
> compression is
> enabled, then sparse files will be allocated and written to
> the disk
I did mention this one in the thread. I personally use lz4 on all
my zpools but I'm happy to see that zstd has been integrated now
as well!
Should have better compression ratios AND better performance.
- John
prev parent reply other threads:[~2021-01-05 16:03 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-04 5:50 NNCP discussion on comp.mail.uucp John Goerzen
2021-01-04 8:36 ` Sergey Matveev
2021-01-04 17:46 ` John Goerzen
2021-01-05 15:07 ` Sergey Matveev
2021-01-05 16:01 ` John Goerzen [this message]