ZeusRAM is STEC's next SSD
PCI product coming
STEC is developing a ZeusRAM non-volatile cache product and a PCIe-connected solid state drive (SSD).
Zeus is STEC's product line for enterprise-class solid state drives (SSD) and the current ZeusIOPS SSD product is used by EMC, IBM, and many others.
Spotted at EMC World, ZeusRAM comes with a standard 8GB of RAM capacity and will be offered as a general cache to STEC's OEM, many of which are sampling the device. It will have faster performance than the ZeusIOPS line as it is a DRAM-based product, not a NAND flash one.
Obviously it has none of the wear-leveling problems characteristic of flash. The company says it is partly non-volatile, because, in the event of a power loss a capacitor facility delivers enough power to destage its contents to flash memory.
The company isn't providing any price indications publicly yet, and we can expect it to be formally announced around the end of June.
STEC's marketing guy at the event, Scott Steltzer, also confirmed that it is developing a PCIe-connected SSD. The firm says there are two basic PCIe SSD card designs; one being a set of flash chips with a driver, taken to mean designs like the ioDrive from Fusion-Io, and the other being a RAID card fitted with flash chips. We think this is like the flash-enhanced RAID adapters from the late and lamented Adaptec RAID operation now being sold to PMC-Sierra.
Steltzer said: "Our approach will be fundamentally different to these two basic designs... being a full-on one. We'll treat the flash as flash. It's going to be significantly different and better than the other products out there today." He wouldn't provide any details about speeds and feeds.
With STEC exhibiting at EMC World and EMC using STEC's ZeusIOPS SSD as its Enterprise flash drive in Symmetrix, CLARiiON and other arrays it might seem likely that the ZeusRAM and the PCIe SSD - ZeusPCI maybe - might feature in EMC's future. ®
STEC 'ZeusRAM' -- predictable, but stupid.
DRAM-based SSD has failed in the marketplace, over and over again for almost 30 years.
It has failed because it is a monumentally stupid idea.
Question: Who in their right mind would spend ridiculous sums on a battery-backed DRAM subsystem and then put it a million miles away from the CPU (in latency terms) -- way the hell on the OTHER side of an ANSI T-10 "block device" storage interface? Especially when they can (a) put the same amount of DRAM in the system itself, (b) do it for 1/10th the cost and (c) access that DRAM 100x faster than is possible going through a storage interface?
Answer: No one is going to do this, because it's just as stupid as it sounds. That's why no DRAM-based SSD technology has made even a tick-turd sized impact on the market.
Larger Question: STEC and the rest of the NAND Flash SSD-hypesters keep telling us that all the performance problems with NAND Flash are going to disappear with the next generation of Flash controllers, FTL software, and new NAND parts. If that's true, why the need for STEC, the undisputed Flash SSD market-leader, to now float a DRAM-based SSD?
Answer: Cause the Flash-fix ain't gonna happen. The dirty secret that all the Flash-hypesters are hiding is that NAND flash parts are getting slower as density increases, and there's not a thing that can be done about it.
A couple of weeks ago a big three-letter IT company put out a service alert to their worldwide systems-engineering staff (paraphrasing):
DO NOT PUT (database) JOURNAL RECEIVERS ON FLASH SSD, THIS WILL SLOW DOWN YOUR SYSTEM (compared to spinning disk).
Imagine that -- Flash SSD will slow down your database to the point you need to move back to spinning disk!
Meanwhile, a big two-letter IT company published a TPC-C benchmark on Flash SSD in April, and the response time ballooned to 3x - 7x higher than the same system built on spinning disk.
I think this is the part of the "hype-cycle" that Gartner calls "The trough of dissillusionment".
This is going to be a deep trough indeed!!!