RAID Controller using memory

[Just D]

All,

What ATA133 RAID controllers using DIMM memory do you know? I'm trying to
look around to find something to improve the speed of my home server that I
use for video rendering. Right now I'm having 4 IBM/Hitachi hard drives,
7200, 250 GB each, working as a RAID. It allows me to get 1 TB space with
100 MBytes/sec up and down, but I guess that the speed could be
significantly improved on the random reading/writing if I use RAID
controller allowing me to instal,l say, couple Gigs of memory that I got
from my old PC. There are two 184 pin DIMMs, 1 GB each. Any ideas? I don't
want to switch to Serial right now because I'll need to buy 4 serial hard
drives and I'm not yet ready for that.

Thanks,
Just D.

 

[Rod Speed]

What ATA133 RAID controllers using DIMM memory do you know? I'm trying to look around to find
something to improve the speed of my home server that I use for video rendering. Right now I'm
having 4 IBM/Hitachi hard drives, 7200, 250 GB each, working as a RAID. It allows me to get 1 TB
space with 100 MBytes/sec up and down, but I guess that the speed could be significantly improved
on the random reading/writing if I use RAID controller allowing me to instal,l say, couple Gigs of
memory that I got from my old PC.

Nope, OS level caching should leave that for dead.

And video rendering does **** all random reading/writing anyway.

There are two 184 pin DIMMs, 1 GB each. Any ideas?

Do it at the OS level.

 

[Arno Wagner]

All,

What ATA133 RAID controllers using DIMM memory do you know? I'm trying to
look around to find something to improve the speed of my home server that I
use for video rendering. Right now I'm having 4 IBM/Hitachi hard drives,
7200, 250 GB each, working as a RAID. It allows me to get 1 TB space with
100 MBytes/sec up and down, but I guess that the speed could be
significantly improved on the random reading/writing if I use RAID
controller allowing me to instal,l say, couple Gigs of memory that I got
from my old PC. There are two 184 pin DIMMs, 1 GB each. Any ideas? I don't
want to switch to Serial right now because I'll need to buy 4 serial hard
drives and I'm not yet ready for that.

Adding RAM to the RAID controller is the most expensive and least
option. Better add RAM to your system and let the OS Buffer/cache
have it.

Arno

 

[Just D]

Adding RAM to the RAID controller is the most expensive and least

option. Better add RAM to your system and let the OS Buffer/cache
have it.

Well, first, if the OS cache is smart enough then yes, but if it's dumb,
then no chances to improve the speed. Second, the 32-bit OS has its own
limits, say 4 GIG RAM, to switch to a 64-bit system is not always
reasonable, the apps are not fully compatible. To switch to Linux - I'm not
having so powerful apps for video editing and rendering. Using RAID with
memory looks like a good alternative especially keeping in mind that this
RAM is free for me, I can upgrade the MB and these two DIMMs are free to go
to any RAID.

Just D.

 

[Arno Wagner]

Well, first, if the OS cache is smart enough then yes, but if it's dumb,
then no chances to improve the speed.

Even Microsoft by now has reasonable OS caches, from what I hear.

Second, the 32-bit OS has its own
limits, say 4 GIG RAM, to switch to a 64-bit system is not always
reasonable, the apps are not fully compatible. To switch to Linux - I'm not
having so powerful apps for video editing and rendering. Using RAID with
memory looks like a good alternative especially keeping in mind that this
RAM is free for me, I can upgrade the MB and these two DIMMs are free to go
to any RAID.

You are aware that typical RAID controllers upport far less than
4GB of RAM?

Arno

 

[Just D]

Even Microsoft by now has reasonable OS caches, from what I hear.

Yes, especially if you correct the cache size and timeouts manually...) The
difference between the disk cache in Win2000 and XP is huge, but the cache
system is far from ideal anyway. Also maybe you noticed that all hard drives
are having their own buffers, not so large, but large enough to optimize the
R/W operations and keep some stuff in memory.

You are aware that typical RAID controllers upport far less than
4GB of RAM?

Sure. That's why I asked about these controllers. I know that some expensive
models working in the servers are using memory modules and that's fine. I
was just asking about these models. I could not write the correct query to
Google to get this list, all my tries were ended with some non-significant
info about computers having RAID and using memory, not memory using raid
controllers. I'm not going to discuss that since the topic is very special
and not all people are aware of this stuff in detail.

Just D.

 

[Arno Wagner]

Yes, especially if you correct the cache size and timeouts manually...) The
difference between the disk cache in Win2000 and XP is huge, but the cache
system is far from ideal anyway. Also maybe you noticed that all hard drives
are having their own buffers, not so large, but large enough to optimize the
R/W operations and keep some stuff in memory.

I know about that. But first the disk caches are typically switched
off by RAID controllers. And second, they do not matter a lot today.
Unless you OS has bad buddering/caching. Since I mostly use Linux (which
has an excellent buffer/cache), I wouldn't know what MS actually has.
But OS buffer/cache is both larger and faster than what a RAID
controller or disk has.

 

[Just D]

Unless you OS has bad buddering/caching. Since I mostly use Linux (which

has an excellent buffer/cache), I wouldn't know what MS actually has.
But OS buffer/cache is both larger and faster than what a RAID
controller or disk has.

True, but this cache doesn't need to be very fast, but it shoul dbe smart
enough to minimize the head movements and that can significantly increase
the R/W speed. Also there are some predictions and if the app reads the same
file again and again the system should be smart enough to read ahead. Most
caches are doing that but the results are different. Also I don't want to
write the content of the cache as soon as I got it causing TRRRRRRRR from
the hard drives, but instead this info should be cached in the buffer to be
flashed from one try. It's not a random read/write operation but a sequence
of very predictable movements and any smart cache system should be able to
take care of that. Also as bigger is the cache as better if we're talking
about these huge files. I realize that if we're talking about a huge number
od pretty small files the cache system can lose some efficience, but that's
another one problem related to the software implementations of these
systems.

So nobody answered the question, but we got a discussion instead...
That's why the newsgroups are misleading sometimes. I didn't ask why and
what for, I just asked what.

Just D.

 

[Folkert Rienstra]

I know about that. But first the disk caches are typically switched
off by RAID controllers. And second, they do not matter a lot today.
Unless you OS has bad buddering/caching. Since I mostly use Linux (which
has an excellent buffer/cache), I wouldn't know what MS actually has.

But OS buffer/cache is both larger

and faster

Nonsense.
Overall the system gets slower by gathering data that may never be used.

 

[Just D]

"Folkert Rienstra"

Nonsense.
Overall the system gets slower by gathering data that may never be used.

http://publib.boulder.ibm.com/infocenter/eserver/v1r2/index.jsp?topic=/diricinfo/fqy0_cadaptiv.html

If you don't know about that it doesn't mean that it doesn't exist

Try to google for it yourself.

http://www.google.com/search?sourceid=navclient&ie=UTF-8&rlz=1T4SKPB_enUS214US214&q=read+ahead+cache

Just D.

 

[Just D]

"Folkert Rienstra"

Nonsense.
Overall the system gets slower by gathering data that may never be used.

Btw, if the hard drive read the cylinder sector by sector the speed would be
muuuuuch slower. Maybe sir knows about low level formatting, physical and
logical geometry of the hard drives, cylinders, sectors, interleave, etc. to
say what he said?

Just D.

 

[Arno Wagner]

"Folkert Rienstra"

Btw, if the hard drive read the cylinder sector by sector the speed would be
muuuuuch slower. Maybe sir knows about low level formatting, physical and
logical geometry of the hard drives, cylinders, sectors, interleave, etc. to
say what he said?

Hehe. Just like Folkert to deny that a perfecly good and established
thing like read-ahead works.

Arno

 

[Kwyjibo]

Overall the system gets slower by gathering data that may never be used.

Oh, that's a keeper.
Do you really believe the absolute shit you post?

 

[Folkert Rienstra]

"Folkert Rienstra"

Btw, if the hard drive read the cylinder sector by sector

Heh, as if there is any other way.

the speed would be muuuuuch slower.

Nope.

Only if the scattered sectors of a noncontiguous file are in a 256 sec-
tor range since that is the maximum of a single read command anyway.
If they are out of that range you have to issue several commands anyway,
whether they are in the same cylinder (track) or not makes no difference.
You'll never manage to read a whole cylinder in a single command.
And whether using several commands to get noncontiguous sectors (clus-
ters actually) is "muuuuuch" slower than a single one is debatable too.
That depends on the command overhead and the minimum read is at least
a cluster worth.

And what has optimizing (ie combining) read commands to do with
read ahead.

Maybe sir knows about low level formatting, physical and logical
geometry of the hard drives, cylinders, sectors, interleave, etc. to
say what he said?

Is that english or you just gibbering.

 

[Folkert Rienstra]

"Folkert Rienstra"

http://publib.boulder.ibm.com/infocenter/eserver/v1r2/index.jsp?topic=/diricinfo/fqy0_cadaptiv.html

If you don't know about that it doesn't mean that it doesn't exist

What has that got to do with the "system".

That's what you get for snipping comments out of context.

Try to google for it yourself.

http://www.google.com/search?sourceid=navclient&ie=UTF-8&rlz=1T4SKPB_enUS214US214&q=read+ahead+cache

I don't care.
The comment was about the system caching -which you conveniently snipped- not controler caching which doesn't involve system memory.

Unless the system caches in idle time, reading more data than is strictly
needed means memory is unavailable longer and the cpu made dependent
on it's own cache to carry on processing. If it runs out it has to wait.
At least this is how it used to be.

Maybe dual port memory controllers are common now, allowing PCI and CPU to access memory at the same time, I'm not up to speed with
CPU chipsets. Google was no help on that and I don't think dual channel me-
mory is the same thing as dual ported memory.