Originally posted by: Idontcare
Originally posted by: IntelUser2000
The Intel drive can actually stutter because its too fast, but it will be very random and sporadic unlike the OCZ drive because the problem is different.
See the Intel drives to Write Combining. Write Combining is using bursts rather than sending data little by little. Write Combining effectively reduces write amplification(the ratio of number of pages/block) because it can burst close to the size of the block. The drive uses log table inside the controller to manage what to do.
Imagine what happens though if the random write speeds greatly exceed the garbage collection speed. If there are enough IOs being fed to the drive that is random writes the lookup table will be optimized for random write speeds and possibly flood the drive because it will become faster.
Of course it can be changed over long time by changing the usage pattern. That sounds similar to what Intel's talking about when saying it optimizes according to usage doesn't it?
Are there any reported observations of this form of performance degradation? (that's not a challenge to your claims, I am curious to see how one characterizes this facet of SSD stutter)
One thing we can count on is being in the murky dark on anything about Intel drives that aren't super
until Intel releases their next-gen drives. Intel likes to demonize their past offerings a little when they release a new product, so you can bet when the new SSD's are released then Intel marketing will start the demonization process on the existing SSD line.
http://forum.ssdworld.ch/viewtopic.php?f=1&t=82
These guys stress SSDs rather hard. The X25-E and the FusionIO which has similar degradations probably wouldn't experience that in normal conditions(at least not that much).
The Write Combining guarantees write amplification to be low if done well but if the other parts of the drive can't keep up slowdowns are possible. I'd think the other way of lowering write amplification is using
speculation. The controller would try to speculate the size of the next data so it can fit on the block. Performance would be higher but there will be a reliability sacrifice.
I'll tell you what happens with my system. I have the X25-M as the main boot drive, and stores most of the applications.
On the Wi-Fi connection which has a strong signal I don't notice anything special.
On the Wi-Max connection(Canadian Bell WiMax) which I have low connection, compared to the Raptor I wouldn't notice much internet lag. But internet lag is of course not about 100% connection. See on the Raptor I'd see constant high latency but on the X25-M I see low latency, and suddenly everything that should have happened will occur after 10-15 sec delay. This happens on WoW and to a less degree on internet and streaming data.
Does it make sense?? Everything floods because the WC algorithm is bursting in large data sizes. In the OCZ SSD it lags consistently and all the time but on the X25-M its sporadic and hard to predict.
Maybe because the X25-M and X25-E has the same controller the X25-M can't stand it. Perhaps the controller is made for X25-E and no special optimization have been made for non-server usage. No personal user needs more than 300-400 IOPS.