SSD - Seq RW Performance Matters

[ksec]

I have posted previously, wondering why SSD based Toshiba controller would perform faster then Sandforce in many cases

http://forums.anandtech.com/showthread.php?t=2116831&highlight=

And when we hit minimal returns with SSD Speed Gains

http://forums.anandtech.com/showthread.php?t=2106805&highlight=

A lot of people, including myself thought Random RW was main factor when we look at SSD performance. And with so many test focusing on it and its importance to our performance Metrics. We all thoughts one day Seq Read Write will get so fast that its speed wouldn't matter and we will focus on Random RW improvement. However it turns out we were wrong.

If the Anand benchmarks are the closes performance matrices we get to show real life SSD performance. Then the recent preview on Vertex 3 just established the fact that having much high Random RW doesn't actually give you much benefits in normal users' real life Applications.

It is the Seq Read Write Speed that matter most on our daily apps performance. And this would explain why Toshiba Controller, despite having a relatively poor performance for Random Read Write, still manage to fair so well in many test due to its faster Seq Read Write performance.

 

[taltamir]

the reason random writes are in such a huge focus is because very early jmicron SSDs had 1/100th the random write speed of a spindle HDD while having better sequential speeds.
And a modern SSD will be over 100x the random write/read speed of a HDD while having 2x the sequential read/write.

100x is a huge MASSIVE difference, in either direction. It doesn't make RW more important, just more noticeable. That being said, we should avoid declaring one or another "important" and look at overall performance which is a blend of all the various speeds the drive can handle.

 

[Zebo]

Which name drives should have Toshiba controller?

 

[ksec]

the reason random writes are in such a huge focus is because very early jmicron SSDs had 1/100th the random write speed of a spindle HDD while having better sequential speeds.
And a modern SSD will be over 100x the random write/read speed of a HDD while having 2x the sequential read/write.

100x is a huge MASSIVE difference, in either direction. It doesn't make RW more important, just more noticeable. That being said, we should avoid declaring one or another "important" and look at overall performance which is a blend of all the various speeds the drive can handle.

Yes, that is why it is Random Read Write that has reach the tipping point, or end of the curve. At least according to today's software development and usage. We need to focus more back on Seq Speed. Because the world seems to be obsessed with Random Read Write.

 

[Rubycon]

Problem is with these disks getting faster we're quickly approaching a brick wall - the ultimate bottleneck of the interface. A faster interface needs to be developed NOW so in the near future these fast(er) disks will not be severely crippled.

 

[ksec]

Problem is with these disks getting faster we're quickly approaching a brick wall - the ultimate bottleneck of the interface. A faster interface needs to be developed NOW so in the near future these fast(er) disks will not be severely crippled.

I was actually thinking the other day Thunderbolt makes a great Internal Interface as well.

It provides enough power for SSD. ( Some are actually approaching 6 - 8 W in operation which sounds ridiculous to me )

Unlike SATA which is Half Duplex, ThunderBolt is Full Duplex.

1Gbps, plus being Full Duplex would gives more then double the bandwidth available compare to SATA 6Gbps.

 

[Rubycon]

I'm surprised they have not made official "SSD Ready" badges for peripherals similar to "Digital Ready" badges for consumer audio gear in the 80s. 😀