Re: [IPCop-user] Will the SATA drivers in IPCop V2.0 support TRIM?

[David W S. <dws...@ov...>]

On 8/18/2011 9:22 AM, Bao Ha wrote:
> On Thu, Aug 18, 2011 at 4:57 AM, ft...@mi...
> <ft...@mi...>wrote:
>
>> I don't want to start a war of words, but just ask if the SATA drivers
>> in the IPCop V2.0 kernel will provide TRIM support.  I'm thinking of
>> building a small low power, fanless IPCop box and a small, recycled SSD
>> would help keep the heat down.  Does anyone know if an SSD is a good idea?
>>
>> Yes and No.
>
> SSD is flash memory and is subject to wear and tear.
>
> To use SSD successfully, you would not put the swap and on it.  You would
> also try not to do caching or proxy.
>
> You would also need to use these options for the root filesystem in the
> /etc/fstab:
>
> "rw,noatime,commit=60"
>
> We calculated that a Kingstom 8GB SSD can survive at least 3 years with a
> 128KB block write (erase) every 45 seconds.  This is for pfSense, but should
> be applicable to IPCop as well.
>

Ok, let me try this again with a real email client:

Actually all one needs to do is choose flash install just prior to 
format during install, it is designed to write to disk only once an 
hour, more on that follows. I have always sworn by flash installs (done 
correctly with mkflash) in 1.4 and 2.0 is improved twofold over the 
standard flash method in 1.4. One, you can choose flash install during 
installation, Two, rather than the RD filesystem that really has not 
been necessary since the 2.2 kernel but was the method in 1.4, flash 
installs  in 2.0 use TMPFS for the ramdisk which is superior in every 
respect to RD. It needs no formatting and can be resized on the fly 
while mounted with no data loss. I also like the low power draw and low 
heat of a flash installs. It's a winner the whole way around if done 
correctly as is the case with flash installs.


The way the flash version works is that /var/log and some of the html 
graphs as well as cron are actually held in a mounted TMPFS (Ramdisk) 
and written to tarballs once an hour for backup ON the disk. These 
tarballs are actually untargzipped on boot as well AFTER the TMPFS is 
mounted as /ram. In the event of a cold reboot or power loss, the most 
logs you will ever lose are close to an hour. I use a UPS and have the 
logs write to my server anyway. Running all the rapidly changing info in 
ram rather than disk and writing to disk only once an hour increases the 
life expectancy of a flash drive to years (15 maybe?) and since you have 
all your graphs and volatile info running in ram rather than on disk, it 
performs extremely well. It also uses NO swap. As far as proxy cache, if 
you want it, give it enough ram for the desired size. I only use the 
proxy for url filtering these days and really need no cache but that's 
my preference.


  I actually switched flash installs to TMPFS in 2008 on my Raqcop 
adaptation of IPCop for Cobalt hardware. I borrowed it from 2.0svn. 
TMPFS also allows you to use a percentage of ram rather than a fixed 
value. I set mine at 50%. Any of that 50% not used by the info in /ram 
in my case is still available to the system to use if need be. I should 
also mention that I am using flash raid on a Cobalt test box and have 
been able to do so for some time with the 2.0svn branch.On a standard PC 
it's like cake.

-- 
Dave Studeman
http:/www.raqcop.com