Summary RZ28M-VW

From: <71055.111_at_compuserve.com>
Date: Tue, 16 Sep 1997 14:45:21 -0400

Hello all,


Thanks to those who replied, especially to:

 Alan <alan_at_nabeth.cxo.dec.com>
 Kurt Carlson <kcarlson_at_arsc.edu>

Kurt Carlson wrote:

Maybe,
If you have a sustained load of 70-ish tps it is probably loaded.
Burst rate (a few snapshots) that's nothing to worry about.

With stripping, your aggregate IO rates can be number of disks * rate
per disk as a peak, reality it will not be quite that high (depends
on mix of the access, whether it's mostly random or sequential).
With mirroring, it's the same as a single disk for writing but
can be faster for reading (typically they use a least used algorithm
for device selection).

Also, keep in mind the documented service times are *averages*,
position of data on the disks can affect that dramatically by
forcing head movements (random vs. sequential access).

Bottom line, if you are consistantly averaging 70tps and the disk
is being used for on-line (vs. batch) type accesses, you may wish
to balance the IO load onto other disks.


Alan wrote:

It is a potentially heavy load. One can get more transfers
        per unit time by using a smaller transfer size and by doing
        strictly sequential transfers. But these are games one
        plays to get the maximum transfer rate and aren't necessarily
        useful. Given the average I/O size around 8 KB, this could
        easily be a heavy load.

        At the same time, you have realize that I/O rate depends on
        the distance the head has to traverse between transfers.
        If the I/O load is multi-threaded and SCSI adapter allows
        command queuing the disk could easily be sorting the disks
        and moving back and forth between the extremes of the capacity
        range. It is possible that it could handle more.

        But, if I had an array that busy, I'd look for ways to move
        some of the load to a different disk/array, get an array with
        more disks, or use faster disks.


My original question:


Hello,

I'm using the following hardware:

Alpha Server 2100, StorageWorks raid array 200, two logical drives: the
first is composed by 2 RZ28D-VW configured to use Raid 1 and the other
composed by 3 RZ28D-VW configured to use Raid 0.

I notice that on the logical drive that is using Raid 0 I'm having the
following average:

Kbs Tps

2151 221

I obtained this statistics using uiao.

It means that on my RZ28D-VW disks I have an average of 73 Tps ( 221/3 ),
RZ28D-VW has an average access time of: 13.5 - 15.1 ms ( I read this on
Internet ). It means that these disks are heavily loaded because the
maximun requests by seconds of them are 66-74 ( 1000/13.5 - 1000/15.1 ).

Is the previous true ?

I will appreciate any comments.

TIA,

Abdon

 
Received on Tue Sep 16 1997 - 21:44:04 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:36 NZDT