Subject: Re: siop(4) and tagged queuing
To: Manuel Bouyer <bouyer@antioche.lip6.fr>
From: Matthew Jacob <mjacob@feral.com>
List: current-users
Date: 10/23/2000 09:58:39
> Hi,
> I just added tagged queuing support to siop(4), and also reworked a bit the
> driver so that it should create less interrupt and PCI load.
> It should also support the QUEUE FULL status (which ncr(4) failed to handle),
> but I didn't find a drive which would generate such a condition. If someone
> has the hardware to test please tell me how it goes.

Almost all of your drives will do this- what you need to do is bump
SDOUTSTANDING up above 100 or so in sd.c.

> 
> I did some bonnie runs on a 4G UW IBM drive attached to a 875 (machine is a
> PII/200):
> -current, with 1.5 siop code:
>               -------Sequential Output-------- ---Sequential Input-- --Random--
>               -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
>            50 11302 92.6 11782 28.1  2960 10.2  8136 86.0 12420 20.4 129.0  2.8
>            50 11228 93.9 12032 30.9  2896 10.1  8454 89.0 12346 20.3 130.8  3.0
>            50 11232 91.5 12042 29.7  2906 10.1  8262 87.5 12542 20.3 131.6  3.0
> 
> -current with new siop code, tagged queueing disabled:
>            50 11286 95.1 11752 28.5  2972 10.4  8300 86.9 12452 19.3 129.7  2.7
>            50 11403 95.8 11964 29.4  2900 10.2  8240 86.8 12496 20.1 134.1  2.8
>            50 11385 95.7 11502 29.0  2934 10.4  8234 86.7 12496 20.2 131.0  2.7
> 
> -current with new siop code, tagged queueing enabled:
>            50  8873 77.2  8908 21.2  2951  9.9  8030 84.6 12420 19.7 129.2  2.6
>            50  8742 75.5  8845 20.7  2889  9.6  8184 86.3 12441 19.9 129.5  2.7
>            50  8791 77.5  8781 21.5  2920  9.6  8175 85.9 12445 19.4 129.2  2.7
> 
> So on average the new code is sighly faster (it should be more obvious
> on a adapter without a internal RAM). However tagged queuing seems to not be
> that good, especially for write - maybe it hurts with the disks' cache.
> I'd be interested in results with others disks :)

You won't see the benefits of tagged queiing on bonnie- you need a
multithreaded test that tests multiple parts of the drive.

-matt