Rob van der Heij wrote:
> Those are orders of binary magnitude?   :-)    And half of that is ?
Ask someone who works on performance. I do never publish measurements
taken on my guest.

> The thing I really like about swapping to DCSS is that you don't drop
> out of SIE and avoid going through CP and be questioned on whether you
> can continue. I would still need a good experiment to show that
> benefit (other than being able to swap faster).
> The official word of Velocity is that we believe the advantage at
> common swap rates do not outweigh the complexity of hacking mem= and
> the risk of getting it wrong on some systems.
The major benefit is _not_ related to direct mem-mem copy or SIE
intercepts. As opposed to the dasd device driver, which is optimized
for real disk IO, we have done quite a few optimizations to our dcss
block device driver that makes it fit perfect for virtual devices:
- no request building/merging
- no device plug mechanism
- no IO schedulers
- no start subchannel and interrupt semantics
- just synchronous memcopy

Carsten
--

Carsten Otte
IBM Linux technology center
ARCH=s390

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to