Re: Running PostgreSQL with ZFS ZIL

From: Vick Khera <vivek(at)khera(dot)org>
To: pgsql-general <pgsql-general(at)postgresql(dot)org>
Subject: Re: Running PostgreSQL with ZFS ZIL
Date: 2015-07-02 13:35:54
Message-ID: CALd+dcdBLodLNdLqkUX0KiNfodQy6i6KXysTQ+udB_PGzP9g=Q@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

On Wed, Jul 1, 2015 at 5:07 PM, Joseph Kregloh <jkregloh(at)sproutloud(dot)com>
wrote:

> We recently built a new server for our Production database. The machine is
> top of the line with 128GB of RAM, dual E5-2650. We also included NVME
> drives for ZIL and L2ARC. Currently we have 3 zpools. First one holds the
> FreeBSD install. Second holds the jails, and third holds all of the
> database data. Needless to say it's fast.
>

FWIW I did not find having a ZIL beneficial for my workload on a similarly
huge servers, also running FreeBSD 10. I do have the L2ARC on the SSD, but
the size of my data set usually leaves the ARC as sufficient to handle
almost all requests. That is, the L2ARC is mostly empty most of the time
(or at least never gets re-fetched from).

I'd start by testing the speed of the driver running the NVME drive. Does
it show up as a normal drive in FreeBSD? I've only ever used regular Intel
SSDs. I don't know if NVME devices connect differently. I hear if you ask
very nicely to the right people you can get special drivers for some fancy
PCI-e based "drives" which make them really fast.

In response to

Responses

Browse pgsql-general by date

  From Date Subject
Next Message Sylvain MARECHAL 2015-07-02 14:11:53 Getting the value of the old_tuple using the test_decoding extension and BDR
Previous Message Lukasz Wrobel 2015-07-02 10:58:18 Re: very slow queries and ineffective vacuum