Semantics of `-Xmn` (Re: hotspot-gc-dev Digest, Vol 186, Issue 12)

[Ramakrishna, Ramki]

Thank you, StefanK, StefanJ, and Thomas, for your detailed responses!

I guess it's complicated ...

Like StefanJ stated, we could fix `-Xms` to state "initial and minimum", and leave `-Xmn` for now, or may be add some verbiage that includes the phrase `best effort` __

Later,
-- Ramki


------------------------------

Message: 2
Date: Tue, 6 Dec 2022 14:47:04 +0100
From: Stefan Karlsson <stefan.karlsson at oracle.com>
To: hotspot-gc-dev at openjdk.org
Subject: Re: Semantics of `-Xmn`
Message-ID: <5daa4918-4255-61a1-3077-17efdf569bb3 at oracle.com>
Content-Type: text/plain; charset=UTF-8; format=flowed

On 2022-12-04 03:38, Ramakrishna, Ramki wrote:
> Question: Can `-Xmn` be interpreted as "fix young gen size to the specified value, not allowing any resizing"?
>
> $ java -X
> ...
>      -Xmn<size>        sets the initial and maximum size (in bytes) of the heap
>                        for the young generation (nursery)
>      -Xms<size>        set initial Java heap size
>      -Xmx<size>        set maximum Java heap size
> ...
>
> Note that this states "sets the initial and maximum size", as though one might be allowed to make the size smaller, but never larger than this.
>
> Similarly `-Xms` sets the initial heap size and `-Xmx` the maximum heap size. Would one be allowed to make the heap size smaller than the `-Xms` specification
>
> It seems as though at least historically one has (implicitly) identified "initial" with "minimum". Thus, we have asked customers to set `-Xms` and `-Xmx` to the same value to prevent changes to the heap size.

FWIW, a few years ago we went over some of the documentation around
`-Xms` to make it clear that that it meant both "initial" and "minimum".
It would probably be good to change the `java -X` output as well.

I'll let others comment about the `-Xmn` flag.

StefanK
...
    ------------------------------

    Message: 3
    Date: Wed, 7 Dec 2022 08:38:06 +0100
    From: Stefan Johansson <stefan.johansson at oracle.com>
    To: Stefan Karlsson <stefan.karlsson at oracle.com>,
            hotspot-gc-dev at openjdk.org
    Subject: Re: Semantics of `-Xmn`
    Message-ID: <f255be0d-e77d-9b4e-7909-ebbfee714296 at oracle.com>
    Content-Type: text/plain; charset=UTF-8; format=flowed

    Hi Ramki,

    On 2022-12-06 14:47, Stefan Karlsson wrote:
    > On 2022-12-04 03:38, Ramakrishna, Ramki wrote:
    >> Question: Can `-Xmn` be interpreted as "fix young gen size to the
    >> specified value, not allowing any resizing"?
    >>

    Kind of, but not exactly for G1. With G1 setting `-Xmn` will create a
    policy that doesn't recalculate the desired size of the young generation
    and sets it to the given value. But this is just a "desired" size and it
    doesn't set a fixed size for the old generation. So if the application
    keeps a lot of objects alive the old generation might grow "too large"
    and cause the young generation to be smaller than what is set with `-Xmn`.

    Same goes for `-XX:NewSize` in G1 it will not be a hard minimum but set
    the desired minimum to this value.

    Given how G1 work I think the below text for `-Xmn` is ok, but I agree
    with StefanK that we should probably update the text for `-Xms`.

    Hope this helps.

    StefanJ
...
    ------------------------------

    Message: 4
    Date: Wed, 7 Dec 2022 09:20:22 +0100
    From: Thomas Schatzl <thomas.schatzl at oracle.com>
    To: hotspot-gc-dev at openjdk.org
    Subject: Re: Semantics of `-Xmn`
    Message-ID: <483b7d69-670a-945a-99c0-d2967efb50ed at oracle.com>
    Content-Type: text/plain; charset=UTF-8; format=flowed

    Hi,

    On 07.12.22 08:38, Stefan Johansson wrote:
    > Hi Ramki,
    >
    > On 2022-12-06 14:47, Stefan Karlsson wrote:
    >> On 2022-12-04 03:38, Ramakrishna, Ramki wrote:
    >>> Question: Can `-Xmn` be interpreted as "fix young gen size to the
    >>> specified value, not allowing any resizing"?
    >>>
    >
    > Kind of, but not exactly for G1. With G1 setting `-Xmn` will create a
    > policy that doesn't recalculate the desired size of the young generation
    > and sets it to the given value. But this is just a "desired" size and it
    > doesn't set a fixed size for the old generation. So if the application
    > keeps a lot of objects alive the old generation might grow "too large"
    > and cause the young generation to be smaller than what is set with `-Xmn`.
    >  > Same goes for `-XX:NewSize` in G1 it will not be a hard minimum but set
    > the desired minimum to this value.

    +1

    G1's policy in this situation is fairly crude: it by default tries to
    keep this free region reserve (G1ReservePercent) to accommodate
    surviving objects in the next gc.

    Since this is a static measure, this is fairly prone to being wrong in
    either direction, either too large (for large heaps) or too small (for
    small heaps), in the best case using a smaller than expected young gen
    (degrading "nicely") or in the worst case running into Full GCs.

    Not much effort has been put into better behavior in this situation.

    Thanks,
       Thomas