Anonymous_User Absent Member.
Absent Member.
2888 views

Question on using PERC 4e/di

Has anyone enabled the writeback feature in netware to work with the
writeback feature on the PERC which is on by default? Looking for some
performance pickups but dont want to risk any data corruption(I do realize
the cache issues if the server crashes)


Labels (2)
0 Likes
17 Replies
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

I have never done any performance tests on this myself, but Novell
claims that enabling write caching on your disk actually slows
performance on NSS volumes because in order to insure that the NSS
journal is always correct, NSS has to do extra work on write caching
devices to make sure the data is actually written before proceeding.

--
Marcel Cox (using XanaNews 1.17.6.5)
0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Yeah I caught that in the docs. I am wondering if I need to set
write-thru on my raid card instead of write-back now. Or if they are 2
totally different parameters.

Marcel Cox wrote:
> I have never done any performance tests on this myself, but Novell
> claims that enabling write caching on your disk actually slows
> performance on NSS volumes because in order to insure that the NSS
> journal is always correct, NSS has to do extra work on write caching
> devices to make sure the data is actually written before proceeding.
>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Jeff Johnson wrote:

> Yeah I caught that in the docs. I am wondering if I need to set
> write-thru on my raid card instead of write-back now. Or if they are
> 2 totally different parameters.


write-back means write caching, e.g. writes go to memory and are
written in the background.
write-though is no write caching. Date is directly written to disk.

--
Marcel Cox (using XanaNews 1.17.6.5)
0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Right,,,I got that. But my question is when netware is set with
write-back off,,,,,,and you have a raid card with write-back on,,,does
that matter? All dell lsi raid cards come default to write-back on.

Marcel Cox wrote:
> Jeff Johnson wrote:
>
>> Yeah I caught that in the docs. I am wondering if I need to set
>> write-thru on my raid card instead of write-back now. Or if they are
>> 2 totally different parameters.

>
> write-back means write caching, e.g. writes go to memory and are
> written in the background.
> write-though is no write caching. Date is directly written to disk.
>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Jeff Johnson wrote:

> and you have a raid card with write-back on,,,does that matter?


I guess that depends on the HAM driver whether it honours the NetWare
set parameter or whether it ignores it only only looks at the BIOS
setting of the controller.

--
Marcel Cox (using XanaNews 1.17.6.5)
0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Yeah,,,,thats the situation I am in.

Marcel Cox wrote:
> Jeff Johnson wrote:
>
>> and you have a raid card with write-back on,,,does that matter?

>
> I guess that depends on the HAM driver whether it honours the NetWare
> set parameter or whether it ignores it only only looks at the BIOS
> setting of the controller.
>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Better be safe than sorry and configure the BIOS and the NetWare set
parameter to be the same. That way, you don't get any ugly surprise.

--
Marcel Cox (using XanaNews 1.17.6.5)
0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Yeah I am thinking that. I wonder how slow my performance will be moving
to write-thru. Guy at dell said it was two different things,,,but I dont
think he knew what he was talking about. To me that netware param and
the raid card param are talking about the exact same thing.

Marcel Cox wrote:
> Better be safe than sorry and configure the BIOS and the NetWare set
> parameter to be the same. That way, you don't get any ugly surprise.
>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Are we talking about "Enable Hardware Write Back" in Monitor -> Server
Parameters -> Disk?
The explanation provided when highlighting this function is "Allows
drivers to utilize hardware Write Back if supported" If I had written
those words I would be saying "If the hardware has the on board cache
and firmware to support this, then the OS will give the data and write
command to the driver to put into the hardware cache".
The second sentence in the explanation says "IO write requests may be
cached and succeeded before data is actually committed to the media." I
had always assumed that meant the driver returns an immediate "success"
to the OS all the while stuffing the data and commands into the disk
hardware cache. (Of course "success" is not returned if the hardware
cache is full at the moment)

If this were true, how can even the most paranoid file system algorithm
in the OS be slowed down? It's being told "success".

Bob

Marcel Cox wrote:
> Jeff Johnson wrote:
>
>
>>Yeah I caught that in the docs. I am wondering if I need to set
>>write-thru on my raid card instead of write-back now. Or if they are
>>2 totally different parameters.

>
>
> write-back means write caching, e.g. writes go to memory and are
> written in the background.
> write-though is no write caching. Date is directly written to disk.
>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

When I was writing this stuff (way way back when) the philosophy was
"take my orders from the OS, then from the hardware. Any conflict would
be resolved in favor of the OS". If the hardware supported it, fine.
If it did not, I was obliged, depending on how sophisticated I (the
driver) was, to emulate the function, or return a message the function
was not available.

In this case, a RAID card with the "Write Back" on, years ago it meant
the function was available. And, unless told otherwise the hardware
BIOS would use it. I would hope when making the Hardware Write Back
setting in NetWare, that NetWare setting would prevail. That's the way
it was when I was writing this stuff, but times, they change.

Bob



Jeff Johnson wrote:
> Yeah I am thinking that. I wonder how slow my performance will be moving
> to write-thru. Guy at dell said it was two different things,,,but I dont
> think he knew what he was talking about. To me that netware param and
> the raid card param are talking about the exact same thing.
>
> Marcel Cox wrote:
>
>> Better be safe than sorry and configure the BIOS and the NetWare set
>> parameter to be the same. That way, you don't get any ugly surprise.
>>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Robert@SchweinA.com wrote:

> If this were true, how can even the most paranoid file system
> algorithm in the OS be slowed down? It's being told "success".


I assume that if the write back caching is completely hidden from the
OS, then it can't cause any slowdown. However if the OS even thinks
write caching might be enabled, then there seems to be a performance
penalty. For instance, there is currently an issue with NSS performance
on OES/Linux with some SATA drives, and that is caused by the OS not
being able to properly control write caching.

--
Marcel Cox (using XanaNews 1.17.6.5)
0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

Yeah,,,I think I am going to change all my LSI cards from write-back to
write-thru. I wont pretend to understand the issues with using write-back
and journaling systems, but have read that it is a problem. Hopefully my
performance wont suffer to much.

"Marcel Cox" <cimetmc@myrealbox.com> wrote in message
news:Y674f.450$8t.283@prv-forum2.provo.novell.com...
> Robert@SchweinA.com wrote:
>
>> If this were true, how can even the most paranoid file system
>> algorithm in the OS be slowed down? It's being told "success".

>
> I assume that if the write back caching is completely hidden from the
> OS, then it can't cause any slowdown. However if the OS even thinks
> write caching might be enabled, then there seems to be a performance
> penalty. For instance, there is currently an issue with NSS performance
> on OES/Linux with some SATA drives, and that is caused by the OS not
> being able to properly control write caching.
>
> --
> Marcel Cox (using XanaNews 1.17.6.5)



0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

We are hobbled because we do not have authoritative documentation about
the driver. Back in the old days <sigh> if the write-back feature was
part of the hardware we would tell our OS to use it. However, some
drives ran so hot that the hardware cache would loose its mind and
corrupt data. After an OS re-install, we would tell the OS to instruct
the driver to not use the write-back functionality. Performance hits
were not as noticeable because the CPU speed <chuckle> and buss speed
were on the same order of magnitude as the disk access times.

I sure hope NW is not trying to second guess the write-back setting we
enable/disable in Monitor. I'm hoping any NSS performance degradation
would be caused by a poor driver returning conflicting information to
the OS causing the OS to be gun-shy and conservative.

Writing good drivers was an art that took months and it was exciting to
see it run the data and instructions at warp speed accurately between
the hardware and the OS. Any more these days the code factories don't
want to spend any more money and time then necessary to get something
out the door so we can debug it in our production environment.

Bob


Jeff Johnson wrote:
> Yeah,,,I think I am going to change all my LSI cards from write-back to
> write-thru. I wont pretend to understand the issues with using write-back
> and journaling systems, but have read that it is a problem. Hopefully my
> performance wont suffer to much.
>
> "Marcel Cox" <cimetmc@myrealbox.com> wrote in message
> news:Y674f.450$8t.283@prv-forum2.provo.novell.com...
>
>>Robert@SchweinA.com wrote:
>>
>>
>>>If this were true, how can even the most paranoid file system
>>>algorithm in the OS be slowed down? It's being told "success".

>>
>>I assume that if the write back caching is completely hidden from the
>>OS, then it can't cause any slowdown. However if the OS even thinks
>>write caching might be enabled, then there seems to be a performance
>>penalty. For instance, there is currently an issue with NSS performance
>>on OES/Linux with some SATA drives, and that is caused by the OS not
>>being able to properly control write caching.
>>
>>--
>>Marcel Cox (using XanaNews 1.17.6.5)

>
>
>

0 Likes
Anonymous_User Absent Member.
Absent Member.

Re: Question on using PERC 4e/di

So are you saying use it or dont use it?

Robert@SchweinA.com wrote:
> When I was writing this stuff (way way back when) the philosophy was
> "take my orders from the OS, then from the hardware. Any conflict would
> be resolved in favor of the OS". If the hardware supported it, fine. If
> it did not, I was obliged, depending on how sophisticated I (the driver)
> was, to emulate the function, or return a message the function was not
> available.
>
> In this case, a RAID card with the "Write Back" on, years ago it meant
> the function was available. And, unless told otherwise the hardware
> BIOS would use it. I would hope when making the Hardware Write Back
> setting in NetWare, that NetWare setting would prevail. That's the way
> it was when I was writing this stuff, but times, they change.
>
> Bob
>
>
>
> Jeff Johnson wrote:
>> Yeah I am thinking that. I wonder how slow my performance will be
>> moving to write-thru. Guy at dell said it was two different
>> things,,,but I dont think he knew what he was talking about. To me
>> that netware param and the raid card param are talking about the exact
>> same thing.
>>
>> Marcel Cox wrote:
>>
>>> Better be safe than sorry and configure the BIOS and the NetWare set
>>> parameter to be the same. That way, you don't get any ugly surprise.
>>>

0 Likes
The opinions expressed above are the personal opinions of the authors, not of Micro Focus. By using this site, you accept the Terms of Use and Rules of Participation. Certain versions of content ("Material") accessible here may contain branding from Hewlett-Packard Company (now HP Inc.) and Hewlett Packard Enterprise Company. As of September 1, 2017, the Material is now offered by Micro Focus, a separately owned and operated company. Any reference to the HP and Hewlett Packard Enterprise/HPE marks is historical in nature, and the HP and Hewlett Packard Enterprise/HPE marks are the property of their respective owners.