Networker

Re: [Networker] Running the library picker/changer on its own SCSI channel?

2003-12-04 06:18:45
Subject: Re: [Networker] Running the library picker/changer on its own SCSI channel?
From: Howard Martin <howard.martin AT EDS DOT COM>
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Date: Thu, 4 Dec 2003 06:18:40 -0500
I'll have a stab at this George.

The number of drives you have on a SCSI port should be predicated by data
rates you expect to see - assuming you are using drive compression then the
data rate to a single drive will depend on compression you get.
eg 4:1 compression on a LTO-1 should give an average data rate of 40Mbytes/s
- of course this is an average and you might find that a fair amount of
data is compressing at twice this for example.
Secondly you need to be clear what the speed of your SCSI adapters are LVD
SCSI doesn't specify the data rate so LVD ultra2 SCSI has a data rate of
80Mbytes/second and LVD ultra3 SCSI 160Mbytes/s - on top of this
manufacturers and techies often use different terms to describe the same
SCSI ( I believe I've seen ultra3 refered to as ultra160 ) so it's worth
double checking what data rate your card can handle.
So in the case of 2 LTO drives with average compression of 4:1 it would on
average not have any problems with 2 drives on 80Mbyte/sec SCSI. In
practice there would be times when both drives want more than half the
bandwidth and data would be held up on the computer this of course would
stop one or both drives from streaming and you then get bitten with the
time taken to reposition the tape in the drive! On the other hand if your
compression average was 1.5:1 then it would seem fairly onlikely that both
drives would want more than half the bandwidth at the same time and all
would be well.
As with lots of backup and restore things your milage may vary!!

With the mixing of drive and robot arms on the same port I'm on less
certain ground.
The robot arm is a slow SCSI device and mixing slow and fast SCSI devices
on a bus inevitably means that traffic to the arm blocks traffic to the
drives for a significant period, this sound slightly dodgy to me as surely
the LVD interface for the robot arm is buffering the data so that the robot
arm speed shouldn't affect the SCSI - but hey what do I know?
The other argument I've heard is that the robot arm tends to generate more
errors than a drive and by isolating it on its own bus stops the
possibility of SCSI errors reseting the bus or downgrading the speed of the
interface ( well I've seen this happen with Suns) both could affect the
data transfer to a drive.

And finally I've just remembered that the minimum number of devices on a
bus gives you less oppurtunity of bringing a bus down if there are drive
problems eg. 2 drives on a bus if one drive dies and needs to be replaced
the other drive will be unavailable whilst the drive is being replaced.

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=