LSI’s Take on Data Center Flash
LSI Corp. has launched a new blog that covers (among other things) flash storage. It’s only natural – the company’s SandForce subsidiary is riding high on the SSD wave and LSI’s HBAs are finding widespread use, both internally and externally, in the production of two-hop PCIe SSDs.
A recent post called “What are the Driving Forces Behind Going Diskless” by LSI Fellow Rob Ober outlines the leading Continue reading
Violin Bets on Both Sides
Violin Memory today made some important announcements. The company has introduced a new line of one-hop PCIe SSDs, and Toshiba will be carrying these as its own products. This creates a tighter link between the two companies: Toshiba is already an investor in Violin, and Toshiba Japan already sells Violin’s memory arrays in Japan.
The new PCIe SSDs are based on Violin’s high-performance NAND management technology, and Violin claims that they offer a higher performance/price point that is available from any other PCIe SSD vendor. This is a key point because PCIe SSD performance varies across a very broad range.
But why does The SSD Guy say that Violin’s betting on both sides? Ever since discrete SSDs started to find their way into the data center there has been a heated debate: Do SSDs belong on the server side of the network or as shared storage? In a virtualized configuration all storage is shared to allow a task to easily move from one server to the next. In an HDD-based system this makes a lot of sense, since an HDD’s latency is significantly larger than that of the network. With SSDs that equation changes – the SSD’s latency is significantly lower than that of the network. The network dramatically reduces the performance of the SSD, so it makes more sense to move that part of the storage into the server, but this breaks the “Shared Storage” model.
This post’s graphic comes from a slide that I have shown repeatedly explaining that it’s wrong to take sides in this argument. Eventually solid state storage will find its way into both sides of the network. In the server it serves as an alternative to large DRAMs, and if it is managed as memory and not as storage, then there will never be any data consistency problems. Of course, most caching solutions also help in this regard, allowing server-side flash to be managed as persistent storage. On the other side of the network flash as shared storage also makes a lot of sense since it accelerates access to shared storage, which is the basis for all virtualized systems.
Violin has taken this argument to heart, betting on flash adoption on both sides of the network, and is the first flash storage array start-up to do so. It will now only be a matter of time before others fall in line.
Objective Analysis covers the enterprise SSD market very closely, issuing reports like our well-regarded annual update covering the enterprise SSD market: The Enterprise SSD: Technologies & Markets. We also perform custom consulting in this area. Clients who wish to engage with us are welcome to drop The SSD Guy a line.
SSDs and TCO
One of the best arguments to use an SSD is also one of the most difficult ways to sell anything. This is the Total Cost of Ownership, commonly abbreviated to “TCO.”
TCO has been used as an argument for buying anything from compact fluorescent bulbs to Jaguar automobiles.
The argument usually revolves around an item whose initial price is higher, but which has lower ongoing (or operating) costs, and when these costs are combined, the higher-priced item proves to cost less to own over the long run. In the case of a compact fluorescent (CF) bulb, the bulb may cost $7, versus $1 for an incandescent bulb, but it consumes 18 Watts compared to the 75 Watts consumed by the incandescent bulb it replaces. In addition the CF bulb lasts ten times as long (10,000 hours vs. 1,000 hours.) This works out to a savings of 470 kWh – or about $50 – plus $3 in bulb costs. Continue reading
‘Noisy Neighbor’ or Cuckoo?
SolidFire has launched a campaign about a phenomenon the company calls the “Noisy Neighbor.” This term is used to express a concept in which one very demanding application absorbs all of the data center’s storage resources to the performance detriment of all other applications. The company points out that this leads to performance variability and poor Quality of Service (QoS.) This, in turn, can drive the enterprise to shun cloud-based services.
The SSD Guy sees this phenomenon as something similar to a “Denial of Service” (DoS) attack, or even the way that cuckoos reproduce. One resource demands more than its fair share of support driving performance below acceptable levels. Heck, even the politics of water rights works this way. In this particular case the constrained resource isn’t network bandwidth, food, or water, but storage bandwidth.
At the bottom of its Noisy Neighbor press release SolidFire has posted an interesting infographic that explains this phenomenon and brings the consequences to monetary terms. Naturally it advocates solid state storage as a solution to the problem. It’s worth a look.
Extreme SSD Error Correction
At last week’s International Solid State Circuits Conference (ISSCC) Shuhei Tanakamaru, a researcher from Japan’s Chuo University, detailed a scheme to reduce MLC SSD bit error rates (BER) by 32 times over conventional techniques. The approach used an impressive combination of mirroring, vertical and horizontal error correction, and a deep understanding of the most likely kinds of bit errors flash will experience.
This is a very novel and well-conceived technique that may find industry adoption in future SSDs.
The steps included in the paper are used in addition to the Continue reading
New Booklet: How Controllers Maximize SSD Life
SNIA (The Storage Networking Industry Association) has conferred a great honor upon the SSD Guy by bringing all of the blog posts in the series How Controllers Maximize SSD Life into a single printed volume of the same name.
Readers can either ask for a print copy from SNIA, or can download a pdf rendition by visiting the SNIA SSSI (Solid State Storage Initiative) education web page.
Seagate Jumps into PCIe SSDs with Virident
Seagate Technology has made a $40 million investment in Virident Systems and has already begun to offer Virident PCIe SSDs to its customers. This was announced on January 28 by the two companies.
Objective Analysis issued an Alert a few hours after the announcement to our clients. Some items we pointed out: Continue reading
The NAND Band!
During this month’s Storage Visions conference, SMART Storage Systems hosted a “NAND Band” party. The company kept the details secret until the guests were all there, after which two “Blues Brothers” impersonators (SMART’s president John Scaramuzzo and Rick Neff, Director of Business Development) showed up in a video singing their new rendition of the 1966 Spencer Davis Group hit: “Gimme Some Lovin’.” SMART’s version was called: “Gimme Some Endurance” and the lyrics centered around the importance of endurance in SSDs.
(SMART’s NAND Band should not be confused with the techno band named NAND which I only discovered while writing this post.)
The reception was held only a couple of hours after Continue reading
Video: How SSDs are Made
Micron Technology has produced a very good four-minute video that shows how SSDs are made. It starts with the design & manufacture of the NAND flash chips themselves and goes right through their assembly into SSDs, interspersing videos of manufacturing processes with graphics of the product’s operation.
This is on the SSD Guy’s recommended YouTube videos list. Check it out by clicking:
How to Locate a Storage Bottleneck
Few Sysadmins really understand what’s happening in the storage interface of their systems, yet there’s a lot of talk about SSDs with wide-ranging IOPS figures along with case studies of how these have helped solve system slowdowns. The big question is: “How do you determine what your storage bottleneck is, and even whether or not one exists at all?”
Tom Coughlin and I discovered a very low level of understanding of this issue when we performed the IOPS survey late last year that we documented in our report: How Many IOPS do You Really Need? A disconcerting number of respondents gave replies that Continue reading
