<div>Thanks Art.</div>
<div>If I'm understanding you correctly, I should check to make sure that
the extra capacity is coming from more stacked platters, and not just more
tracks. ( that is probably the case ).</div>
<div>And even then I will suffer some small decrease in performance as
tables fragment. </div>
<div> </div>
<div>But, it is very im****tant to continue to put data on these larger
disks by striping it the way we have been, with raid and plaid.</div>
<div> </div>
<div>Is that about what you said ?</div>
<div> </div>
<div>Thanks !!
<br>
<br>
<br>
<br>----- Original Message -----
<br>Subject: Re: disk drive selection question
<br>From: "Art Kagel" <art.kagel@[EMAIL PROTECTED]
>
<br>Date: Thu, May 8, 2008 9:00
<br></div>
<table>
<tbody>
<tr>
<td width="1" bgcolor="#a2b6d2"> </td>
<td valign="top">
<div class="PrivateMsgDiv"><!-- begin sanitized html -->
<div class="moz-text-html" lang="x-unicode">Floyd,
<br>
<br>Here's the deal. First, sustained read/write speeds assume
sequential IO so the width of the platters does not affect that spec
regardless. Second, I assume these are all 3.5'' or all 5.0'' drives,
not different sizes. For the most part, the drive manufacturers increase
the capacity of their drives by adding more platters not be increasing the
number of tracks, so increased capacity in the same drive product line
(say Seagate Barracuda drives) will all have the same basic performance
characteristics. Here's where there will be a difference between using
36G drives and 146G drives. If you are replacing 4 or 5 striped 36G
spindles with a single 146G spindle to get the same capacity, then there
will be a 75-80% reduction in overall performance because you are only
reading or writing that 58-96mb/sec from/to one spindle instead of from/to
4 or 5 spindles at once. For a single small write that would hit only one
drive in the array you won't see any difference, but
for larger IOs, like at checkpoint time or when sequential scanning a
large table, the difference will be HUGE.
<br>
<br>If you are, on hte other hand, moving from say 5 36G drives to 5 146G
drives in the same stripe then no, there will not be any performance
penalty to get the 4 or 5-fold increase in storage capacity under normal
cir***stances. The only caviat is that when your tables become
fragmented, the detrimental effect can be a bit greater than on the
smaller drives over time simply because there will necessarily be more
head movement to gather the data from the larger drives since there will
be more nooks and crannies for the extra extents to hide in.
<br>
<br>Art S. Kagel
<br>Oninit
<br>
<br>
<br>
<div style="margin-left: 40px;">We are currently running OLTP on and Ibm
P570 aix 5.3 attached to a DS4500 san. We currently have 36g 15k drives,
Fiber Channel disks in there.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">We are running out of space, and drawers.
Our sales people are telling us that we should move to 146g 15k rpm
drives.</div>
<div style="margin-left: 40px;">The supposed specs on those drives, from
an ibm white paper are:</div>
<div style="margin-left: 40px;">3.5ms to read, 4ms to write, max sustained
transfer rate 58-96mb/sec.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">The funny thing is, they have the same
performance rates whether the drive is 73g or 146g.</div>
<div style="margin-left: 40px;">We have been under the impression that the
bigger a drive is, the slower the average IO will be, due to the fact that
the head has to do more moving around to get a bigger part of the
data.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">So my questions basically are:</div>
<div style="margin-left: 40px;">1) Is our premise correct, or is IBM
correct in saying that we will not suffer a performance decrease by moving
to bigger drives ?</div>
<div style="margin-left: 40px;">2) Has anyone done any testing on these ?
I wouldn't be sure where to begin to do my own benchmarking, without
having a totally isolated system. Or am I just reading too much into this
and should just buy the bigger disks.</div>
<div style="margin-left: 40px;">3) What am I missing here ?</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">Any advice is appreciated.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">Thanks,</div>
<div style="margin-left: 40px;">Floyd</div>
<div style="margin-left: 40px;"> </div></div><!-- end sanitized html
--></div></td></tr></tbody></table>
------=_20080508081644_40996
Content-Type: multipart/alternative;
boundary="----=_20080508081644_34668"
------=_20080508081644_34668
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 8bit
Thanks Art.
If I'm understanding you correctly, I should check to make sure that the
extra capacity is coming from more stacked platters, and not just more
tracks. ( that is probably the case ).
And even then I will suffer some small decrease in performance as tables
fragment.
But, it is very im****tant to continue to put data on these larger disks by
striping it the way we have been, with raid and plaid.
Is that about what you said ?
Thanks !!
----- Original Message -----
Subject: Re: disk drive selection question
From: "Art Kagel" >;art.kagel@[EMAIL PROTECTED]
>
Date: Thu, May 8, 2008 9:00
Floyd,
Here's the deal. First, sustained read/write speeds assume sequential IO
so the width of the platters does not affect that spec regardless.
Second, I assume these are all 3.5'' or all 5.0'' drives, not different
sizes. For the most part, the drive manufacturers increase the capacity
of their drives by adding more platters not be increasing the number of
tracks, so increased capacity in the same drive product line (say Seagate
Barracuda drives) will all have the same basic performance
characteristics. Here's where there will be a difference between using
36G drives and 146G drives. If you are replacing 4 or 5 striped 36G
spindles with a single 146G spindle to get the same capacity, then there
will be a 75-80% reduction in overall performance because you are only
reading or writing that 58-96mb/sec from/to one spindle instead of from/to
4 or 5 spindles at once. For a single small write that would hit only one
drive in the array you won't see any difference, but for
larger IOs, like at checkpoint time or when sequential scanning a large
table, the difference will be HUGE.
If you are, on hte other hand, moving from say 5 36G drives to 5 146G
drives in the same stripe then no, there will not be any performance
penalty to get the 4 or 5-fold increase in storage capacity under normal
cir***stances. The only caviat is that when your tables become
fragmented, the detrimental effect can be a bit greater than on the
smaller drives over time simply because there will necessarily be more
head movement to gather the data from the larger drives since there will
be more nooks and crannies for the extra extents to hide in.
Art S. Kagel
Oninit
We are currently running OLTP on and Ibm P570 aix 5.3 attached to a DS4500
san. We currently have 36g 15k drives, Fiber Channel disks in there.
We are running out of space, and drawers. Our sales people are telling us
that we should move to 146g 15k rpm drives.
The supposed specs on those drives, from an ibm white paper are:
3.5ms to read, 4ms to write, max sustained transfer rate 58-96mb/sec.
The funny thing is, they have the same performance rates whether the drive
is 73g or 146g.
We have been under the impression that the bigger a drive is, the slower
the average IO will be, due to the fact that the head has to do more
moving around to get a bigger part of the data.
So my questions basically are:
1) Is our premise correct, or is IBM correct in saying that we will not
suffer a performance decrease by moving to bigger drives ?
2) Has anyone done any testing on these ? I wouldn't be sure where to
begin to do my own benchmarking, without having a totally isolated system.
Or am I just reading too much into this and should just buy the bigger
disks.
3) What am I missing here ?
Any advice is appreciated.
Thanks,
Floyd
------=_20080508081644_34668
Content-Type: text/html; charset="utf-8"
Content-Transfer-Encoding: 8bit
<div>Thanks Art.</div>
<div>If I'm understanding you correctly, I should check to make sure that
the extra capacity is coming from more stacked platters, and not just more
tracks. ( that is probably the case ).</div>
<div>And even then I will suffer some small decrease in performance as
tables fragment. </div>
<div> </div>
<div>But, it is very im****tant to continue to put data on these larger
disks by striping it the way we have been, with raid and plaid.</div>
<div> </div>
<div>Is that about what you said ?</div>
<div> </div>
<div>Thanks !!
<br>
<br>
<br>
<br>----- Original Message -----
<br>Subject: Re: disk drive selection question
<br>From: "Art Kagel" <art.kagel@[EMAIL PROTECTED]
>
<br>Date: Thu, May 8, 2008 9:00
<br></div>
<table>
<tbody>
<tr>
<td width="1" bgcolor="#a2b6d2"> </td>
<td valign="top">
<div class="PrivateMsgDiv"><!-- begin sanitized html -->
<div class="moz-text-html" lang="x-unicode">Floyd,
<br>
<br>Here's the deal. First, sustained read/write speeds assume
sequential IO so the width of the platters does not affect that spec
regardless. Second, I assume these are all 3.5'' or all 5.0'' drives,
not different sizes. For the most part, the drive manufacturers increase
the capacity of their drives by adding more platters not be increasing the
number of tracks, so increased capacity in the same drive product line
(say Seagate Barracuda drives) will all have the same basic performance
characteristics. Here's where there will be a difference between using
36G drives and 146G drives. If you are replacing 4 or 5 striped 36G
spindles with a single 146G spindle to get the same capacity, then there
will be a 75-80% reduction in overall performance because you are only
reading or writing that 58-96mb/sec from/to one spindle instead of from/to
4 or 5 spindles at once. For a single small write that would hit only one
drive in the array you won't see any difference, but
for larger IOs, like at checkpoint time or when sequential scanning a
large table, the difference will be HUGE.
<br>
<br>If you are, on hte other hand, moving from say 5 36G drives to 5 146G
drives in the same stripe then no, there will not be any performance
penalty to get the 4 or 5-fold increase in storage capacity under normal
cir***stances. The only caviat is that when your tables become
fragmented, the detrimental effect can be a bit greater than on the
smaller drives over time simply because there will necessarily be more
head movement to gather the data from the larger drives since there will
be more nooks and crannies for the extra extents to hide in.
<br>
<br>Art S. Kagel
<br>Oninit
<br>
<br>
<br>
<div style="margin-left: 40px;">We are currently running OLTP on and Ibm
P570 aix 5.3 attached to a DS4500 san. We currently have 36g 15k drives,
Fiber Channel disks in there.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">We are running out of space, and drawers.
Our sales people are telling us that we should move to 146g 15k rpm
drives.</div>
<div style="margin-left: 40px;">The supposed specs on those drives, from
an ibm white paper are:</div>
<div style="margin-left: 40px;">3.5ms to read, 4ms to write, max sustained
transfer rate 58-96mb/sec.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">The funny thing is, they have the same
performance rates whether the drive is 73g or 146g.</div>
<div style="margin-left: 40px;">We have been under the impression that the
bigger a drive is, the slower the average IO will be, due to the fact that
the head has to do more moving around to get a bigger part of the
data.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">So my questions basically are:</div>
<div style="margin-left: 40px;">1) Is our premise correct, or is IBM
correct in saying that we will not suffer a performance decrease by moving
to bigger drives ?</div>
<div style="margin-left: 40px;">2) Has anyone done any testing on these ?
I wouldn't be sure where to begin to do my own benchmarking, without
having a totally isolated system. Or am I just reading too much into this
and should just buy the bigger disks.</div>
<div style="margin-left: 40px;">3) What am I missing here ?</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">Any advice is appreciated.</div>
<div style="margin-left: 40px;"> </div>
<div style="margin-left: 40px;">Thanks,</div>
<div style="margin-left: 40px;">Floyd</div>
<div style="margin-left: 40px;"> </div></div><!-- end sanitized html
--></div></td></tr></tbody></table>
------=_20080508081644_34668--
------=_20080508081644_40996--
|
