[prev in list] [next in list] [prev in thread] [next in thread] 

List:       tux-list
Subject:    Re: Keyed cache
From:       Marcus Gruendler <runner () tamalin ! de>
Date:       2001-02-11 23:07:08
[Download RAW message or body]

On Sunday 11 February 2001 19:30, you wrote:
[...]
> you should use the GET_OBJECT/SEND_OBJECT combo to 'simulate' a static
> reply - this will do the right async-IO thing even if the cached reply is
> no longer in memory and must be read from disk. You can use normal file
> I/O to create new files for TUX to serve if you want to cache dynamic
> replies. (In fact natively accelerated static replies are just a
> GET_OBJECT + SEND_OBJECT done internally within the kernel.)
>
> the performance hit to go over user-space is not significant. I did some
> measurements on a 550 MHz PIII box (1 processor used), serving a 50 kbyte
> static reply (non-keepalive):
>
>  - natively accelerated static reply: 1210 replies/sec (495 Mbits)
>  - user-space module static reply:    1120 replies/sec (458 Mbits)
>
[...]

Ingo, could you post the code of the user-space module you used in this 
benchmark? 

Since I am still a little confused about how to use the async event mechanism 
of TUX while using the GET_OBJECT/SEND_OBJECT actions, such an example would 
be _very_ usefull to me :)

Bye, Marcus

-- 
Marcus Gruendler
eMail: runner@tamalin.de
WWW: http://www.tamalin.de/runner/index.html



_______________________________________________
tux-list mailing list
tux-list@redhat.com
https://listman.redhat.com/mailman/listinfo/tux-list

[prev in list] [next in list] [prev in thread] [next in thread]