 | |
#21
 
| |||
| |||
|
Re: Raid configurations -
12-27-2008
, 07:43 AM
|
"Bernd Felsche" <berfel (AT) innovative (DOT) iinet.net.au> wrote: "S W" <SW (AT) spam (DOT) invalid> wrote: How important is RAID 10 on a Progress 9.1E installation? |
Quote:
|
You need to do a detailed, device and file-level IO analysis to identify the particular bottleneck and add bandwidth at that point. Without an analysis, you'll be throwing a lot of time and money at the problem, trying to fix it, with only a random chance that it will actually fix it; and an even better chance that it will make things worse. RAID is not a panacea. RAID is more of a placebo. :-) Since posting the original question, we have done detailed IO analysis using different "models" or scenarios of data server uses, and found that the system is actually OK at reading data, but maybe a little slow at writing it. I now agree that RAID 10 would not make a huge difference. But its still slow. We're having a Progress analyst doing a system health check in the next few days, which hopefully will throw more light onto things. |
Depending on where you are writing, more spindles and spreading the
data and meta-data can make a big difference.
Non-volatile RAM drives (Flash) now seem to be getting there; with
durability better than that of rotating media. Data density isn't
quite so good, nor purchase cost.
Progress RDBMS systems would benefit most dramatically by putting
the BI on "solid-state disk". If your databases a small enough, you
can whack the whole lot onto one card. (Keep the HDD for AI and
backups.)
Fusion-IO are about to start shipping (finger crossed) PCI cards
with sustained write speeds of 600 megabytes per second (100,000+ IO
per second). There is no seek latency. It is about 100 times faster
than a 10,000 rpm HDD. The card plugs into a PCI-E slot, bypassing
the HDD hardware.
<http://www.dvnation.com/Fusion-IO-IODrive-SSD-Solid-State-Disk-Drive-Review.html>
It looks very promising. Revolutionary and paradigm-shifting.
But Santa didn't leave one for me at Christmas. :-(
2009 is likely to see the first NUMA-extended servers on the market;
with slots for such flash-based non-volative RAM as well as normal
DRAM. JEDEC are discussing standards. Hope they won't be slowed too
much by the financial crunch.
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | Second to agriculture, humbug is the
X against HTML mail | biggest industry of our age.
/ \ and postings | -- Alfred Nobel
#22
| |||
| |||
|
|
"Bernd Felsche" <berfel (AT) innovative (DOT) iinet.net.au> wrote: "S W" <SW (AT) spam (DOT) invalid> wrote: How important is RAID 10 on a Progress 9.1E installation? |
Quote:
|
You need to do a detailed, device and file-level IO analysis to identify the particular bottleneck and add bandwidth at that point. Without an analysis, you'll be throwing a lot of time and money at the problem, trying to fix it, with only a random chance that it will actually fix it; and an even better chance that it will make things worse. RAID is not a panacea. RAID is more of a placebo. :-) Since posting the original question, we have done detailed IO analysis using different "models" or scenarios of data server uses, and found that the system is actually OK at reading data, but maybe a little slow at writing it. I now agree that RAID 10 would not make a huge difference. But its still slow. We're having a Progress analyst doing a system health check in the next few days, which hopefully will throw more light onto things. |
Depending on where you are writing, more spindles and spreading the
data and meta-data can make a big difference.
Non-volatile RAM drives (Flash) now seem to be getting there; with
durability better than that of rotating media. Data density isn't
quite so good, nor purchase cost.
Progress RDBMS systems would benefit most dramatically by putting
the BI on "solid-state disk". If your databases a small enough, you
can whack the whole lot onto one card. (Keep the HDD for AI and
backups.)
Fusion-IO are about to start shipping (finger crossed) PCI cards
with sustained write speeds of 600 megabytes per second (100,000+ IO
per second). There is no seek latency. It is about 100 times faster
than a 10,000 rpm HDD. The card plugs into a PCI-E slot, bypassing
the HDD hardware.
<http://www.dvnation.com/Fusion-IO-IODrive-SSD-Solid-State-Disk-Drive-Review.html>
It looks very promising. Revolutionary and paradigm-shifting.
But Santa didn't leave one for me at Christmas. :-(
2009 is likely to see the first NUMA-extended servers on the market;
with slots for such flash-based non-volative RAM as well as normal
DRAM. JEDEC are discussing standards. Hope they won't be slowed too
much by the financial crunch.
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | Second to agriculture, humbug is the
X against HTML mail | biggest industry of our age.
/ \ and postings | -- Alfred Nobel
#23
| |||
| |||
|
|
"Bernd Felsche" <berfel (AT) innovative (DOT) iinet.net.au> wrote: "S W" <SW (AT) spam (DOT) invalid> wrote: How important is RAID 10 on a Progress 9.1E installation? |
Quote:
|
You need to do a detailed, device and file-level IO analysis to identify the particular bottleneck and add bandwidth at that point. Without an analysis, you'll be throwing a lot of time and money at the problem, trying to fix it, with only a random chance that it will actually fix it; and an even better chance that it will make things worse. RAID is not a panacea. RAID is more of a placebo. :-) Since posting the original question, we have done detailed IO analysis using different "models" or scenarios of data server uses, and found that the system is actually OK at reading data, but maybe a little slow at writing it. I now agree that RAID 10 would not make a huge difference. But its still slow. We're having a Progress analyst doing a system health check in the next few days, which hopefully will throw more light onto things. |
Depending on where you are writing, more spindles and spreading the
data and meta-data can make a big difference.
Non-volatile RAM drives (Flash) now seem to be getting there; with
durability better than that of rotating media. Data density isn't
quite so good, nor purchase cost.
Progress RDBMS systems would benefit most dramatically by putting
the BI on "solid-state disk". If your databases a small enough, you
can whack the whole lot onto one card. (Keep the HDD for AI and
backups.)
Fusion-IO are about to start shipping (finger crossed) PCI cards
with sustained write speeds of 600 megabytes per second (100,000+ IO
per second). There is no seek latency. It is about 100 times faster
than a 10,000 rpm HDD. The card plugs into a PCI-E slot, bypassing
the HDD hardware.
<http://www.dvnation.com/Fusion-IO-IODrive-SSD-Solid-State-Disk-Drive-Review.html>
It looks very promising. Revolutionary and paradigm-shifting.
But Santa didn't leave one for me at Christmas. :-(
2009 is likely to see the first NUMA-extended servers on the market;
with slots for such flash-based non-volative RAM as well as normal
DRAM. JEDEC are discussing standards. Hope they won't be slowed too
much by the financial crunch.
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | Second to agriculture, humbug is the
X against HTML mail | biggest industry of our age.
/ \ and postings | -- Alfred Nobel
#24
| |||
| |||
|
|
"Bernd Felsche" <berfel (AT) innovative (DOT) iinet.net.au> wrote: "S W" <SW (AT) spam (DOT) invalid> wrote: How important is RAID 10 on a Progress 9.1E installation? |
Quote:
|
You need to do a detailed, device and file-level IO analysis to identify the particular bottleneck and add bandwidth at that point. Without an analysis, you'll be throwing a lot of time and money at the problem, trying to fix it, with only a random chance that it will actually fix it; and an even better chance that it will make things worse. RAID is not a panacea. RAID is more of a placebo. :-) Since posting the original question, we have done detailed IO analysis using different "models" or scenarios of data server uses, and found that the system is actually OK at reading data, but maybe a little slow at writing it. I now agree that RAID 10 would not make a huge difference. But its still slow. We're having a Progress analyst doing a system health check in the next few days, which hopefully will throw more light onto things. |
Depending on where you are writing, more spindles and spreading the
data and meta-data can make a big difference.
Non-volatile RAM drives (Flash) now seem to be getting there; with
durability better than that of rotating media. Data density isn't
quite so good, nor purchase cost.
Progress RDBMS systems would benefit most dramatically by putting
the BI on "solid-state disk". If your databases a small enough, you
can whack the whole lot onto one card. (Keep the HDD for AI and
backups.)
Fusion-IO are about to start shipping (finger crossed) PCI cards
with sustained write speeds of 600 megabytes per second (100,000+ IO
per second). There is no seek latency. It is about 100 times faster
than a 10,000 rpm HDD. The card plugs into a PCI-E slot, bypassing
the HDD hardware.
<http://www.dvnation.com/Fusion-IO-IODrive-SSD-Solid-State-Disk-Drive-Review.html>
It looks very promising. Revolutionary and paradigm-shifting.
But Santa didn't leave one for me at Christmas. :-(
2009 is likely to see the first NUMA-extended servers on the market;
with slots for such flash-based non-volative RAM as well as normal
DRAM. JEDEC are discussing standards. Hope they won't be slowed too
much by the financial crunch.
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | Second to agriculture, humbug is the
X against HTML mail | biggest industry of our age.
/ \ and postings | -- Alfred Nobel
#25
| |||
| |||
|
|
"Bernd Felsche" <berfel (AT) innovative (DOT) iinet.net.au> wrote: "S W" <SW (AT) spam (DOT) invalid> wrote: How important is RAID 10 on a Progress 9.1E installation? |
Quote:
|
You need to do a detailed, device and file-level IO analysis to identify the particular bottleneck and add bandwidth at that point. Without an analysis, you'll be throwing a lot of time and money at the problem, trying to fix it, with only a random chance that it will actually fix it; and an even better chance that it will make things worse. RAID is not a panacea. RAID is more of a placebo. :-) Since posting the original question, we have done detailed IO analysis using different "models" or scenarios of data server uses, and found that the system is actually OK at reading data, but maybe a little slow at writing it. I now agree that RAID 10 would not make a huge difference. But its still slow. We're having a Progress analyst doing a system health check in the next few days, which hopefully will throw more light onto things. |
Depending on where you are writing, more spindles and spreading the
data and meta-data can make a big difference.
Non-volatile RAM drives (Flash) now seem to be getting there; with
durability better than that of rotating media. Data density isn't
quite so good, nor purchase cost.
Progress RDBMS systems would benefit most dramatically by putting
the BI on "solid-state disk". If your databases a small enough, you
can whack the whole lot onto one card. (Keep the HDD for AI and
backups.)
Fusion-IO are about to start shipping (finger crossed) PCI cards
with sustained write speeds of 600 megabytes per second (100,000+ IO
per second). There is no seek latency. It is about 100 times faster
than a 10,000 rpm HDD. The card plugs into a PCI-E slot, bypassing
the HDD hardware.
<http://www.dvnation.com/Fusion-IO-IODrive-SSD-Solid-State-Disk-Drive-Review.html>
It looks very promising. Revolutionary and paradigm-shifting.
But Santa didn't leave one for me at Christmas. :-(
2009 is likely to see the first NUMA-extended servers on the market;
with slots for such flash-based non-volative RAM as well as normal
DRAM. JEDEC are discussing standards. Hope they won't be slowed too
much by the financial crunch.
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | Second to agriculture, humbug is the
X against HTML mail | biggest industry of our age.
/ \ and postings | -- Alfred Nobel
 |
« Previous Thread
|
Next Thread »
| Thread Tools | |
| Display Modes | |
Linear Mode
| |
Powered by vBulletin Version 3.5.3
Copyright ©2000 - 2013, Jelsoft Enterprises Ltd.
Copyright ©2000 - 2013, Jelsoft Enterprises Ltd.