Quote:
Source:  PC Magazine, April 20, 2007, page 122

What do the members of this forum know about HHDs?  My understanding is that they are essentially a standard hard disk drive, but contain an embedded flash memory drive on which Windows Vista can maintain its SuperFetch (ReadyDrive) cache.  Why this would be faster than the Windows Vista ReadyBoost feature (an external USB flash drive used as a cache) is unclear to me, however.

Thanks for the note, MrMagoo.

To summarize, SuperFetch (ReadyDrive) transfers data from flash memory via the SATA interface of the hard disk drive in which it is embedded, whereas ReadyBoost transfers data from flash memory via the USB interface – and, the latter is slower than the former.

Now I see.

Another interesting article from Lenovo to add to the mix.

Some points I'd add:
- You don't want writes buffered in a removable cache, so hybrids get half a tick for write buffering. Having the solid-state memory integrated elsewhere in the system where it's non-removable is a different matter, as noted in Lenovo's article.
- Integrated caches fill themselves parasitically; external caches need to be filled separately, which incurs a cost. The cost may be small, but it needs to be close to zero for the cache to be worth filling speculatively.
- Existing hybrid drives get pretty small caches; this is for low-cost applications. For performance applications, you want to be using a quality RAID controller anyway, and that's a much more intelligent place to put some solid-state storage than on the drives, because it can be a lot bigger.

One of the developers here at work recently build himself a new home gaming system and dropped a hefty chunk of change on a 4-port 3Ware PCIe RAID controller. Yesterday his opinion was that the 3Ware was well worth every dollar, and made more difference than any other component.

It's early days, though. I can see some species of flash cache becoming effectively universal in notebooks for the power-management benefits alone, and that means that the long-run manufacturing costs of each type are going to end up playing a big part; Microsoft pushing hybrids as part of co-marketing with Vista probably means that they'll get to enjoy real enconomies of scale sooner than other approaches, which could seal the deal.

Heh, good point there Rad. Mind you, it's interesting to see how SCSI and IDE have evolved into SAS and SATA; it's particularly interesting that they engineered SAS to be able to host SATA drives as well.

We have a few xSeries 226 IBM boxen downstairs in the QA lab, and they are pretty nice machines (especially for the price), loaded up with a couple SAS drives they crank pretty well - and they throw some SATA connectors into the mix as well so you can add boatloads of commodity storage as well.

I think they took your guide to heart . In terms of sheer flexibility in storage choices from fast to cheap they are really impressive.

Quote:


I didn't know you could run SATA drives off a SAS controller. Guess the operative word is "serial".

Any particular controllers/adapters or brands you like?

Rad wrote on Apr 9^th, 2007 at 9:52am:

I only know about what we see come through QA's hardware purchases, and they are more about general compatibility testing rather than performance benchmarking. The xSeries 226's were bought last year specifically to test SAS (a few being kitted with dual EMT64 Xeons as platform tests for x64 Server 2003, which did incidentally reveal a difference for our code compared to the multicore Athlons). So far I don't believe we've looked at separate SAS controllers or other SAS server models; we've got various other lower-end server hardware mostly from Dell and HP, but as lower-end models they tend to have onboard SATA-based RAID.

As to performance, the SATA-to-SAS process is just a tunneling of the SATA command set (much the same as ATAPI was just a tunneling of the original SCSI command set over an ATA command), so not much difference there. However, the host controllers in the 226's are quite different from each other (a RAID controller versus a low-overhead Intel AHCI SATA host adapter) making it not entirely apples-to-apples.

Still, my gut feeling is that SAS and SATA are pretty close to parity right now; SAS has been planned with a lot more headroom, but right now today both interfaces deliver plenty of bandwidth relative to the sustained drive throughputs - the change to PCIe has also levelled things out a bit (compared to recent years with the server-only halfway-house of PCI-X for add-on host controllers).

Basically, you can build some serious storage arrays out of either. A 3Ware or Areca (check out this review) RAID controller with dedicated hardware XOR and those 750Gb Seagates is going to have terrifyingly high throughput and terrifyingly huge capacity for surprisingly little outlay, while an array of 15,000RPM SAS drives will be smaller and much more expensive but will still naturally win on seek-intensive workloads (manufacturers will probably keep drives in that category SAS much as they did in past generations, as a useful way to distinguish them and thus preserve a price premium from commodity drives).

It's not really an either-or question, but more one of designing storage to match a workload and a budget. There are corners of the price/latency/capacity/throughput space where each technology still has a unique place, and a huge middle where either can deliver well, and where the distinctions come down more to how individual products in the lineup perform than technologies.

A good article on the subject of SuperFetch and ReadyBoost:  Windows Vista's SuperFetch and ReadyBoost Analyzed