DPDK usage discussions
 help / color / mirror / Atom feed
From: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>
To: "Lombardo, Ed" <Ed.Lombardo@netscout.com>
Cc: users <users@dpdk.org>
Subject: Re: DPDK 22.11.2 requires too much VIRT memory, how to reduce
Date: Fri, 10 Nov 2023 12:37:39 +0300	[thread overview]
Message-ID: <20231110123739.21fbcb01@sovereign> (raw)
In-Reply-To: <20231110123133.2adbd2cc@sovereign>

2023-11-10 12:31 (UTC+0300), Dmitry Kozlyuk:
> Hi Ed,
> 
> 2023-11-10 00:16 (UTC+0000), Lombardo, Ed:
> > I finally finished testing all the options and found the VIRT value can be
> > reduced from 66 GB to 16 GB with --legacy-mem setting in EAL init argument.  
> 
> Right.
> By default, DPDK can use up to 64 GB of hugepage memory,
> so it reserves 64 GB of VIRT (but does not map most of it),
> RES should be low until the app actually allocates something.
> In legacy mode, DPDK maps all available hugepage memory at startup,
> in your case 16 GB, VIRT and RES should be close.
> 
> >  So I therefore had to increase the VM memory from 16 GB to 24 GB (instead
> > of 80 GB without this setting).  
> 
> I don't understand why you have to do that.
> Possible VIRT is not limited by available RAM.
> DPDK should be able to reserve 64 GB of VIRT on a machine with 16 GB of RAM,
> it will just be unable to map more than 16 GB (obviously).

Sorry, I've sent the message early by mistake.

> > I wonder what do we give up with this setting?

Most importantly, in legacy mode DPDK will consume all available hugepages at
startup and will not free them back to the system until the all is terminated.
The default dynamic mode allocates and frees physical RAM on demand.
Some advanced DPDK memory API don't work in legacy mode.

> > 
> > All the other settings I tried and combinations of these had no impact
> > (socket-limit=2048, single-file-segments, no-shconf, and no-telemetry) on
> > VIRT memory.

Right, they should not.
DPDK assumes that VIRT reservation is almost free and unlimited.
May it be that your system somehow limits it?





  reply	other threads:[~2023-11-10  9:37 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-11-05 16:00 Lombardo, Ed
2023-11-05 18:22 ` Dmitry Kozlyuk
2023-11-08 17:48   ` Lombardo, Ed
2023-11-08 18:12     ` Dmitry Kozlyuk
2023-11-08 22:26       ` Lombardo, Ed
2023-11-08 23:05         ` Dmitry Kozlyuk
2023-11-10  0:16           ` Lombardo, Ed
2023-11-10  9:31             ` Dmitry Kozlyuk
2023-11-10  9:37               ` Dmitry Kozlyuk [this message]
2024-03-14  2:51                 ` Lombardo, Ed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20231110123739.21fbcb01@sovereign \
    --to=dmitry.kozliuk@gmail.com \
    --cc=Ed.Lombardo@netscout.com \
    --cc=users@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).