[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] Ovewriting Checkpoint platform



And there's the piece of the puzzle I suspected that I was missing! Thanks Greg!

It sounds like the substring-based platform matching -- where an ssse3 executable would be able to resume on an "ssse3 ssse4_1 ssse4_2 machine, but not vice-versa -- would be the proper way to go for the original poster, then.

        -Michael Pelletier.

"HTCondor-users" <htcondor-users-bounces@xxxxxxxxxxx> wrote on 10/12/2015 11:32:43 AM:

> From: Greg Thain <gthain@xxxxxxxxxxx>

> To: htcondor-users@xxxxxxxxxxx
> Date: 10/12/2015 11:33 AM
> Subject: Re: [HTCondor-users] Ovewriting Checkpoint platform
> Sent by: "HTCondor-users" <htcondor-users-bounces@xxxxxxxxxxx>
>
> The reason HTCondor seems a bit more picky here than you might expect is
> that many applications use SSE instructions unknowingly -- several of the
> glibc string functions detect if the target machine has SSE once at
> program startup, and then use string functions optimized for that
> particular sub-model for the duration of the program run.  If a restored
> checkpoint starts up on a different submodel, the program usually crashes.
>
> -greg_______________________________________________