Re: Debian not suitable for SSD due to apt/dpkg?
On 2012-10-01 13:32, Frank Bauer wrote:
> On Mon, Oct 01, 2012 at 10:23:32AM +0800, Chow Loong Jin wrote:
>>
>> Have you done any actual calculation on this? A quick Google search on SSD write
>> cycles shows more articles debunking this theory than supporting it.
>
> Reading specifications of intel's SSD 320 line at the following link:
>
> http://www.intel.com/content/www/us/en/solid-state-drives/ssd-320-enterprise-server-storage-application-specification-addendum.html
>
> states in section 2.3 Reliability that for 160GB drive the write
> endurance is 15TB, which gives about 94 full overwrite cycles. Not
> that much in my eyes.
And it will take about 70 days to do these 15 TB of 4KB writes
distributed over the full device (600 IOPS according to Table 1)
(15*10^12/4096/600/3600/24)
while 5000 sequential overwrites could be finished within 53 days
(5000*160*10^9/165/2^20/3600/24) (165 MB/s, Table 2)
Of course you may not interrupt your write workload by reading, which
would delay the wearout.
Do you have any information about the overprovisioning and erase block
size of that device?
> The point, however, is not whether the SSD in question will last 94 or
> 100 overwrites, but whether Debian's packaging system is not causing
> too much unneccessary write overhead.
Neither of the above extremes will probably match *your* workload.
I would assume apt-get update will mostly do sequential writes (on
several files) and this will cause some file system meta-data updates
which may result in "more random" I/Os (but restricted to a small
portion of the disk) and they go to the filesystem log, too.
That specification also gives some nice instructions how to use SMART
attributes to analyze your work load ... did you look into this?
Andreas
Reply to: