Mailing List Archive

Hi,

just read this: http://www.blachford.info/computer/articles/bigcrunch1.html

It's about the future of processors, and the problems that lie ahead in
multicore design. I am glad we are designing Varnish with this in mind
since singlethread processes could end up meet themselvs in the door,
literally.

Anders Berg

In message <1706.193.213.34.102.1146516990.squirrel at denise.vg.no>, "Anders Berg
" writes:

>just read this: http://www.blachford.info/computer/articles/bigcrunch1.html

I must admit that I find this entire "multi-everything" to be a bit,
all-right: very, overhyped these days.

Writing programs which scale to just 32 cores is incredibly hard,
unless we talk about something like CGI scripts in apache2 (and
even then it's not trivial.

At the bottom of it is the fact that Intel has promised to double
performance 18 months, at least until Moore croaks.

Having run up against the 4GHz Si/CMOS clock barrier, the only way out
is expanding number of cores.

But in reality, the silicon real-estate could be used much better if
you take a "whole system performance" view.

For instance: add hw-crypto (Intel/AMD doesn't dare do this, because
they would have to make sure CPU's don't get into Iraq, N.Korea etc
but VIA already did it.)

--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

In message <1706.193.213.34.102.1146516990.squirrel at denise.vg.no>, "Anders Berg
" writes:

>just read this: http://www.blachford.info/computer/articles/bigcrunch1.html

I must admit that I find this entire "multi-everything" to be a bit,
all-right: very, overhyped these days.

Writing programs which scale to just 32 cores is incredibly hard,
unless we talk about something like CGI scripts in apache2 (and
even then it's not trivial.

At the bottom of it is the fact that Intel has promised to double
performance 18 months, at least until Moore croaks.

Having run up against the 4GHz Si/CMOS clock barrier, the only way out
is expanding number of cores.

But in reality, the silicon real-estate could be used much better if
you take a "whole system performance" view.

For instance: add hw-crypto (Intel/AMD doesn't dare do this, because
they would have to make sure CPU's don't get into Iraq, N.Korea etc
but VIA already did it.)

--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

> -----Opprinnelig melding-----
> Fra: varnish-dev-bounces at projects.linpro.no
> [mailto:varnish-dev-bounces at projects.linpro.no] P? vegne av
> Poul-Henning Kamp
> Sendt: 2. mai 2006 08:47
> Til: andersb at vgnett.no
> Kopi: varnish-dev at projects.linpro.no
> Emne: Re: Predicting the future
>
> In message
> <1706.193.213.34.102.1146516990.squirrel at denise.vg.no>,
> "Anders Berg " writes:
>
> >just read this:
> >http://www.blachford.info/computer/articles/bigcrunch1.html
>
> I must admit that I find this entire "multi-everything" to be a bit,
> all-right: very, overhyped these days.

I agree, unfortunately AMD and Intel follow/lead the hype to some degree.

> Writing programs which scale to just 32 cores is incredibly
> hard, unless we talk about something like CGI scripts in
> apache2 (and even then it's not trivial.

Yes, thats why the T1 is good on some benchmarks I understand. And, if I am not mistaking, also Varnish will be?

Anyway, this article gave me (probably not you guys) an easy-to-understand glimpse of the possible future.

> At the bottom of it is the fact that Intel has promised to
> double performance 18 months, at least until Moore croaks.
>
> Having run up against the 4GHz Si/CMOS clock barrier, the
> only way out is expanding number of cores.
>
> But in reality, the silicon real-estate could be used much
> better if you take a "whole system performance" view.

Yeah, but apps need to be aware of this right? I think Varnish is in a better position than the "competition" to take this into account. :)

Anders Berg

> For instance: add hw-crypto (Intel/AMD doesn't dare do this,
> because they would have to make sure CPU's don't get into
> Iraq, N.Korea etc but VIA already did it.)
>
> --
> Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
> phk at FreeBSD.ORG | TCP/IP since RFC 956
> FreeBSD committer | BSD since 4.3-tahoe
> Never attribute to malice what can adequately be explained by
> incompetence.
> _______________________________________________
> varnish-dev mailing list
> varnish-dev at projects.linpro.no
> http://projects.linpro.no/mailman/listinfo/varnish-dev
>


*****************************************************************
Denne fotnoten bekrefter at denne e-postmeldingen ble
skannet av MailSweeper og funnet fri for virus.
*****************************************************************
This footnote confirms that this email message has been swept by
MailSweeper for the presence of computer viruses.
*****************************************************************

> -----Opprinnelig melding-----
> Fra: varnish-dev-bounces at projects.linpro.no
> [mailto:varnish-dev-bounces at projects.linpro.no] P? vegne av
> Poul-Henning Kamp
> Sendt: 2. mai 2006 08:47
> Til: andersb at vgnett.no
> Kopi: varnish-dev at projects.linpro.no
> Emne: Re: Predicting the future
>
> In message
> <1706.193.213.34.102.1146516990.squirrel at denise.vg.no>,
> "Anders Berg " writes:
>
> >just read this:
> >http://www.blachford.info/computer/articles/bigcrunch1.html
>
> I must admit that I find this entire "multi-everything" to be a bit,
> all-right: very, overhyped these days.

I agree, unfortunately AMD and Intel follow/lead the hype to some degree.

> Writing programs which scale to just 32 cores is incredibly
> hard, unless we talk about something like CGI scripts in
> apache2 (and even then it's not trivial.

Yes, thats why the T1 is good on some benchmarks I understand. And, if I am not mistaking, also Varnish will be?

Anyway, this article gave me (probably not you guys) an easy-to-understand glimpse of the possible future.

> At the bottom of it is the fact that Intel has promised to
> double performance 18 months, at least until Moore croaks.
>
> Having run up against the 4GHz Si/CMOS clock barrier, the
> only way out is expanding number of cores.
>
> But in reality, the silicon real-estate could be used much
> better if you take a "whole system performance" view.

Yeah, but apps need to be aware of this right? I think Varnish is in a better position than the "competition" to take this into account. :)

Anders Berg

> For instance: add hw-crypto (Intel/AMD doesn't dare do this,
> because they would have to make sure CPU's don't get into
> Iraq, N.Korea etc but VIA already did it.)
>
> --
> Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
> phk at FreeBSD.ORG | TCP/IP since RFC 956
> FreeBSD committer | BSD since 4.3-tahoe
> Never attribute to malice what can adequately be explained by
> incompetence.
> _______________________________________________
> varnish-dev mailing list
> varnish-dev at projects.linpro.no
> http://projects.linpro.no/mailman/listinfo/varnish-dev
>


*****************************************************************
Denne fotnoten bekrefter at denne e-postmeldingen ble
skannet av MailSweeper og funnet fri for virus.
*****************************************************************
This footnote confirms that this email message has been swept by
MailSweeper for the presence of computer viruses.
*****************************************************************