Breaking News
View Comments: Newest First | Oldest First | Threaded View
DMcCunney
User Rank
Author
Re: It is rightly pitched to the servers market
DMcCunney   12/23/2013 11:34:51 AM
NO RATINGS
@Kinnar: but in case of a normal general purpose server everything goes on one location, in that case it will be required to consider the write endurance.

A normal, general purpose server has only one location?  I doubt it. 

And even if it is on the read-optimized SSD, see my comments about write limits on NAND flash.  How long do you think it's likely to take before write endurance becomes a concern?  Offhand, a rather long time.

I'd be concerned about timely updates, but that would likely be a matter of caching and asynchronous batch writes.

Kinnar
User Rank
Author
Re: Cloud Era
Kinnar   12/23/2013 5:53:03 AM
NO RATINGS
This is a very good solution!!, but separating the write and read operations in cloud environment will be less feasible but playing with local operating systems stack it will be possible to separate the write and read locations on different hard disk drives. But this will be matter of some research and development.

 

Kinnar
User Rank
Author
Re: It is rightly pitched to the servers market
Kinnar   12/23/2013 5:36:14 AM
NO RATINGS
But when a data is read from a server which is defined as read intensive server, it simultaneously requires to write something on the server that will be the log statistics keeping or updates of the content, yes for a database warehouse there will be different locations for storing this data, but in case of a normal general purpose server everything goes on one location, in that case it will be required to consider the write endurance.

DMcCunney
User Rank
Author
Re: It is rightly pitched to the servers market
DMcCunney   12/22/2013 3:42:29 PM
NO RATINGS
For NAND flash, I believe the current limits are abourt 100,000 writes per cell before it goes bad.  And the controller circuitry is designed to transparently move data from failing cells, mark them bad, abd remap, so that what you see is graceful degradation as total drive capacity decreases.  In most cases, I'd expect the flash drive to be removed and replaced with a bigger, faster, better performing unit before degradation is even noticeable.  (That's 100,000 writes per cell. How long will it take for any particular cell to be written to 100,000 times?  How many cells are in a drive?  How long will it take for wear to be noticeable?  Offhand, a long time.)

Because of that, if I'm a server admin, I'm less concerned with write endurance than write speed.

Unless you are doing OLTP with lots of database updates, reading the data quickly will be far more important than writing it, so flash optimized for fast reads can be attractive.

 

z80man
User Rank
Author
Re: It is rightly pitched to the servers market
z80man   12/22/2013 1:46:07 PM
NO RATINGS
For read intensive operations the number of writes that will be needed would be far less that typical and the drive would probably become obsolete before reaching the point of failure. The bigger problem I can see if with the focus of read performance and possible less write performance. Performing updates to files could cause the writes to interfere with the high performance reads.

chanj0
User Rank
Author
Cloud Era
chanj0   12/11/2013 5:45:45 PM
NO RATINGS
There is a write before there is something to read. However, no doubt, an article on the web probably are read a million time while there is only 1 write. The benefit of read fast will definitely improve user experience. What's the speed improvement vs the write penalty?

Depending on the penalty, when a web service is being development with both read and write, one of the many challenges lays on how a cloud system is designed so that read and write are going to different servers.

Kinnar
User Rank
Author
It is rightly pitched to the servers market
Kinnar   12/11/2013 2:32:34 PM
NO RATINGS
This will be really a good storage solution for the read intensive servers, but the write endurance will be again one point for the administrators to think of before accepting the device.

Most Recent Comments
michigan0
 
SteveHarris0
 
realjjj
 
SteveHarris0
 
SteveHarris0
 
VicVat
 
Les_Slater
 
SSDWEM
 
witeken
Most Recent Messages
9/25/2016
4:48:30 PM
michigan0 Sang Kim First, 28nm bulk is in volume manufacturing for several years by the major semiconductor companies but not 28nm FDSOI today yet. Why not? Simply because unlike 28nm bulk the LDD(Lightly Doped Drain) to minimize hot carrier generation can't be implemented in 28nm FDSOI. Furthermore, hot carrier reliability becomes worse with scaling, That is the major reason why 28nm FDSOI is not manufacturable today and will not be. Second, how can you suppress the leakage currents from such ultra short 7nm due to the short channel effects? How thin SOI thickness is required to prevent punch-through of un-dopped 7nm FDSOI? Possibly less than 4nm. Depositing such an ultra thin film less then 4nm filum uniformly and reliably over 12" wafers at the manufacturing line is extremely difficult or not even manufacturable. If not manufacturable, the 7nm FDSOI debate is over!Third, what happens when hot carriers are generated near the drain at normal operation of 7nm FDSOI? Electrons go to the positively biased drain with no harm but where the holes to go? The holes can't go to the substrate because of the thin BOX layer. Some holes may become trapped at the BOX layer causing Vt shift. However, the vast majority of holes drift through the the un-dopped SOI channel toward the N+Source,...

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed