Ext3 users - Feb 2002 - [Off-topic] Battery Backed NVRAM for journals ...

I've seen this come up on occassion, but every NAS OEM that uses
NVRAM I've ever talked to over the last 18 months won't tell me
anything about their equipment nor their suppliers.  Recently, Micro
Memory contacted me.  Is anyone here using their products?

FYI, their 64-bit PCI 128MB-1GB NVRAM board is here:
http://www.micromemory.com/newwebsite/dynamic/content/product.asp?itemID=MM-5415CN

-- Bryan

-- 
Bryan J. Smith, Engineer        mailto:b.j.smith@ieee.org   
AbsoluteValue Systems, Inc.     http://www.linux-wlan.org
SmithConcepts, Inc.          http://www.SmithConcepts.com
---------------------------------------------------------
1999 IRS Data:  The richer half of all US income
                earners pay 99% of all US taxes

"Bryan J. Smith" wrote:
> 
> I've seen this come up on occassion, but every NAS OEM that uses
> NVRAM I've ever talked to over the last 18 months won't tell me
> anything about their equipment nor their suppliers.
A fair bunch of people did a fair bunch of testing of this last year.
Bottom line: RAM-based journals don't speed ext3 up.

The testing was performed with a ramdisk-based journal,
`trd' which is buried in the ext3 CVS server somewhere.  I'd
suggest that you have a good play with that before shelling
out, or getting your hopes up.

Please bear in mind that the external journal support is
completely experimental, may change any time, may rot your
socks, etc.

The disappointing performance changes are not really surprising.
A journal write is a single seek.  And with most any workload,
there's a complete storm of seeking going on, so the cost of
the journal seek is very low.

-

Andrew Morton wrote:
> A fair bunch of people did a fair bunch of testing of this last year.
> Bottom line: RAM-based journals don't speed ext3 up.
> The testing was performed with a ramdisk-based journal,
> `trd' which is buried in the ext3 CVS server somewhere.  I'd
> suggest that you have a good play with that before shelling
> out, or getting your hopes up.
> Please bear in mind that the external journal support is
> completely experimental, may change any time, may rot your
> socks, etc.
> The disappointing performance changes are not really surprising.
> A journal write is a single seek.  And with most any workload,
> there's a complete storm of seeking going on, so the cost of
> the journal seek is very low.
Was this using full-data journaling?
Testing NFS v3 sync clients with large writes?
More importantly, what about post-crash recovery, especially with
NFS?

Just curious.  I could see how this stuff isn't very "cool" for
Samba and typical services, but for NFS v3 servers using sync
writes, I would be shocked if it didn't help at all.

Thanx for the info.  Any URLs to the archives would be greatly
appreciated.

-- Bryan

-- 
Bryan J. Smith, Engineer        mailto:b.j.smith@ieee.org   
AbsoluteValue Systems, Inc.     http://www.linux-wlan.org
SmithConcepts, Inc.          http://www.SmithConcepts.com
---------------------------------------------------------
1999 IRS Data:  The richer half of all US income
                earners pay 99% of all US taxes

On Tuesday February 12, b.j.smith@ieee.org wrote:
> I've seen this come up on occassion, but every NAS OEM that uses
> NVRAM I've ever talked to over the last 18 months won't tell me
> anything about their equipment nor their suppliers.  Recently, Micro
> Memory contacted me.  Is anyone here using their products?
> 
> FYI, their 64-bit PCI 128MB-1GB NVRAM board is here:
>
http://www.micromemory.com/newwebsite/dynamic/content/product.asp?itemID=MM-5415CN
I have one of these.  I did a bit of work on the Linux driver too and I
know a fair bit about it.

I had to send it back last year for a couple of mods and have't had a
chance to test it out again since I got it back, but it looks quite
good.

Used as an external data=journal journal for ext3 it makes sync NFS go
quite nicely.  I haven't yet done a controlled comparison of NFS over
ext3 with journal on one of these versus the same but with the journal
on a mirrored pair of 15000rpm scsi drives, but I suspect it should go
a bit faster.  I'll let you know when I do it.

NeilBrown

> Finally, however, we have to do journal replay --- looking for the
> most uptodate copies of each block in the journal and writing them
> back to disk.  This is again a sequential scan through the journal,
> but the writebacks are NOT sequential and can cause a lot of seeking.
> I suspect that that is where the bulk of the time is spent.
What about sorting those writebacks before commiting them to disk? Or
should the disksubsystem take care of that trough that elevator-alg.?


===============================================De informatie opgenomen in dit
bericht kan vertrouwelijk zijn en
is uitsluitend bestemd voor de geadresseerde. Indien u dit bericht 
onterecht ontvangt, wordt u verzocht de inhoud niet te gebruiken en 
de afzender direct te informeren door het bericht te retourneren. 
===============================================The information contained in this
message may be confidential
and is intended to be exclusively for the addressee. Should you 
receive this message unintentionally, please do not use the contents 
herein and notify the sender immediately by return e-mail.