Articles: Storage

Bookmark and Share

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 ]

Database Patterns

In the Database pattern the drive is processing a stream of requests to read and write 8KB random-address data blocks. The ratio of read to write requests is changing from 0% to 100% with a step of 10% throughout the test while the request queue depth varies from 1 to 256.

You can click this link to view the tabled results for the IOMeter: Database pattern.

We will build diagrams for request queue depths of 1, 16 and 256.

We’ve got two leaders at the minimum queue depth. The second-generation X25-M is a considerable improvement over its predecessor and is absolutely unrivalled at high percentages of reads. The Vertex Mac Edition is ahead of the others at writing. There must be some fundamental changes in its firmware or some other reason, but the other Indilinx-based models behave differently in this test and deliver much lower performance. Interestingly, the Agility hardly differs from its opponents under such loads.

The Summit is a disappointment once again. It is competing with the other OCZ drives at high percentages of reads but slows down awfully when there is some share of writes to be performed.

When the queue depth is increased to 16 requests, Intel’s second-generation drive goes even further ahead at high percentages of reads. It is two times as fast as its closest pursuer! Just think about it: this SSD can process 25 thousand requests to read 8KB data blocks in a single second! Before the arrival of solid state drives such performance could only be achieved by creating virtual disks in system RAM. Even the iRAM device, based on DDR SDRAM, was only half as fast.

There is nothing particularly interesting at the longer queue depth. We still have two leaders at different types of load, the others pursuing them more or less successfully.

Winding up this section of the review we will build diagrams showing graphs for five different queue depths for each SSD.

Intel’s engineers have done a good job writing firmware for their second-generation SSDs. The increased efficiency of NCQ algorithms is obvious. The newer model is obviously faster. We have doubts about whether this is true NCQ, though. The SSD can hardly reorder requests according to their LBA addresses because this would produce no performance benefits. We guess that it reorders requests in such a way as to load all the controller channels as much as possible simultaneously. It is also good that the SSD does not slow down too much at very high percentages of writes.

The OCZ Summit’s diagram resembles the one of the Corsair P128. This platform seems to have a high degree of repeatability. We can see that the SSD gets faster at reading when the queue depth is increased, but suffers a terrible performance hit at writing. There is one difference from the Corsair P128, though. The OCZ Summit successfully copes with the queue depth of 4 requests. It looks like the SSD runs out of buffer memory for write requests at long queue depths.

OCZ’s Agility, Vertex and Vertex Turbo behave in a similar manner. They enjoy a small performance growth when the request queue is increased, but only at reading. The Vertex Turbo differs from the others with its repeatable and inexplicable performance jump at about 70% writes.

The OCZ Vertex Mac Edition differs from the other Indilinx-based products, making us suspect that it uses some secret controller. Its performance increases at longer queue depths even at high percentages of writes. More importantly, this SSD is generally better than its mates at writing. Our praise goes to the developers of its firmware if this is due to the firmware. It is a fantastic result for a drive based on MLC memory to deliver over 4 thousand operations per second under any load!

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 ]


Comments currently: 3
Discussion started: 10/14/09 12:57:27 PM
Latest comment: 10/30/09 06:10:38 PM

View comments

Add your Comment