* Posts by gcn

2 publicly visible posts • joined 23 May 2011

Intel penetrates PCIe flash biz with long-lasting hardness

gcn

It’s important to note that many enterprise applications are looking for high random read/write performance rather than just streaming data performance. For example, many SSD deployments are for OLTP database environments which are heavily oriented towards mixed random read/write I/Os. I think an important question to address is the Intel 910’s random read/write performance on a 4K/8K block size and how that compares to other SSDs (especially when the drive is near full capacity), because that is the performance metric that matters most to the majority of enterprise environments and applications that are currently using SSDs.

TMS launches RamSan-70 Gorilla

gcn

Onboard processor has drawbacks...

TMS architecture’s impact will be felt when the card begins to get pushed (device fully seasoned, full capacity utilization, multiple threads etc.). That’s where Virident tachIOn shines.

Ramsan is able to produce decent micro-benchmarks. But, the strength highlighted in the article: "The RamSan-70 does not rely on the host processor for flash translation as there is an on-board PowerPC CPU" -- is also its weakness.

This means that real-world application performance, where multiple cores are all accessing the flash in parallel, will be difficult to scale. The speed of the PowerPC CPU cannot keep up with multiple cores running at 2.5+ Ghz all pushing I/O in parallel. Straight micro-benchmarks create close comparisons, but the ‘offload’ architectures typically fall behind as soon as you put them into real-world applications.

Disclaimer: I am an employee of Virident!