Frequently Asked Questions

Q: Who benefits from Solid State Disks?

A: Companies who wish to extend the life of their existing infrastructure by increasing performance with an SSD and delaying or possibly avoiding the need to upgrade CPUs and network.

Companies, who wish to improve total response time, batch run times, transactional processing or other applications where there are simultaneous processes with random access to various files.

Applications that require random access to data incur delays due to mechanical disk rotation and head movement. Disk drive vendors and manufacturers of disk array products add cache memory in order to minimize these delays and to attempt to predict what data the host system will require next.

Applications that perform predictive access to data all to often fail to achieve the expected benefit from cache usage, especially at times of peak load when it is most needed, because there are too many simultaneous requests for data from different users. This results in the data that was predicted to be required, often being lost before the host requests it; so it has to be read again, resulting in another mechanical delay. This often leads to the situation where under moderate loads the system response is acceptable but under peak loads it appears to the users that the system has 'ground to a halt'.

In both cases, caches are not totally effective as they are typically small in comparison to the size of the underlying storage and are usually split into a fairly small number of 'segments' (8 for a typical disk drive). As a result even if the data required is of only a moderately random nature the caches all too often provide little or no benefit. With Solid State technology there are no mechanical delays so the time taken to access the data is consistent and very small (usually less than that of a cache hit on a conventional drive) regardless of where on the device the data is held.

Q: What Applications are Good Candidates for SpeedStor?

A: Applications that create a large number of simultaneous, small block, random reads and writes will benefit from Solid State Drives. Typically there are a small number of files that are frequently accessed (20% of the files are accessed 80% of the time).

1. Clustered File Systems: Place centralized Metadata onto SSD. (Example: Sun’s SAM-QFS)

2. Virtual Desktop Infrastructure (VDI): VDI solutions can leverage SpeedStor for their centralized application repositories. For example VirtualStorm has all applications packaged and placed on a centralized disk, where many concurrent users can access one copy of an application.

3. Common Enterprise Workloads: File Server, Mail Server, Voice Mail Server, News Server, Web Server: Although the amount of data is normally quite small these workload see performance improvements because user data is no longer scattered all over the storage device.

4. Trading, Banking, Purchasing, Online transactions, Securities: SSDs decrease access times to the data table portion of the database solid state disks can make a vast difference to the access times to the database index and log files. Solid-state disks improve response times and process more transactions than hard disk drive based storage systems. Compare SpeedStor’s 52,000 I/O's per sec to hard disk based systems that can only achieve 250 I/O’s per sec.

5. Search Engine cache, Swap Files: Solid State disks can also be deployed to speed up systems, which maintain swap files or search engine caches. In these cases they act as a very large data cache that does not have to be reloaded when the system is restarted.

6. Streaming Online Video: SpeedStor will offer blazing performance of online video when the I/O stream is served in multiple 64k paths and with a 0-2m.s. latency.

Q: Are DRAM based systems faster than NAND based systems?

A: This is true, however the question is how fast is fast enough? First of all, recognize that the SpeedStor is the only NAND SSD that can compete in the enterprise market. It is designed to sustain high IOPs (52,000IO’s per second/per drive); whereas the rest of the NANAD SSDs in the marketplace sustain between 2,000 and 7,500 IOPS/drive. Furthermore, there is no other NAND drive that can sustain 18,000 random writes.

The high end DRAM SSDs, like RAMSan are designed with faster chips as well as multiple controllers and are inherently faster. That being said, one often will not see a difference between the RAMSan and SpeedStor due to the fact that bottlenecks are a moving target. Once we improve IO, we increase CPU utilization and bandwidth on the FC network and ports. Benchmarks found that the SpeedStor compared to the DRAM based RAMSan 320 is comparable in performance in a number of environments. It is, however, outperformed by the RAMSan by as much as a factor of 2 when in a read/write profile. What this means is that we know we can improve a customers’ application by (rule of thumb) 10X with the SpeedStor.

If I want to go to 20X then you need the following:

1. A server that can sustain more IO’s (many of our tests showed that both the SpeedStor and the RAMSan drove the CPU to 90%+).

2. An application to sustain the load (90% of our customers will see the “biggest bang” with 10X)

3. Four times the money (RAMSan is 4X the cost) or More ports and more money (place multiple SpeedStor drives on own channel).

4. To weigh the benefits of NAND vs. DRAM in regards to TCO. SpeedStor reduces cost of power utilization and long-term maintenance (five year warranty…no moving parts!).

Q: Are NAND based systems inherently less reliable and put data at risk?

A: There are pros and cons to both NAND and DRAM chip technology. In regards to reliability, it all has to do with how the chips are integrated into the disk drive design. There are various NAND implementations. The SpeedStor is in a class by itself.

The Benefits of NAND vs. DRAM as it applies to reliability:

The known limitation of NAND is the fact that the chips have a limitation of how often one can write to a single cell on the chip. This is presently 100,000 times. The SpeedStor, is designed to avoid this limitation by implementing a high end wear leveling technique. The drive itself has the ability to consistently write (7 x24x365) for 11 to 20 years (depending upon which capacity drive you choose) before one will begin to witness write errors due to the 100K write/cell limitation. This will be further explained below and under “FAQ5”.

Let us first compare The SpeedStor to a typical DRAM SSD implementation (for instance; many of our customers are familiar with RAMSan by Texas Memory). The SpeedStor is a non-volatile disk drive, which means that if power is lost the data stays in tact. The drive itself has the ability to consistently write (7 x24 x 365) for 11 to 20 years (depending upon which capacity drive you choose) before one will begin to witness write errors due to the 100K write/cell limitation. At this point one still does not lose data; one receives an error notice. Given this, a SpeedStor with a single drive is more reliable than a DRAM based SSD. DRAM is volatile. Hence SSD manufacturers must build a battery to backup system to support the DRAM in the event of a power outage, or else all data is lost. Now there are more parts (chips, battery, rotating disk) and more points of failure.

Presently a single 146GB SpeedStor retails for $10,000 (with enclosure that will grow to four drives). SDS recommends mirroring, via disk suite, if one is to place critical data (such as metadata) onto a drive. It is still affordable to implement a high availability solution with the SpeedStor. Our solutions range between $18k (mirrored 146GB SSDs) and $32k (four drives in a 1U-two 146GB SSDs mirrored to an additional two). Whereas it would cost $100+K to mirror a RAMSan…and this is the only way to compare reliability in an apples to apples configuration.

Q: Is it true that NAND SSDs are not suitable for write intensive applications?

A: These calculations are based on the assumption that when you write a part of a file you are changing the same physical media location every time you write. This assumption is not the case. Even if the same host block is written over and over the physical storage of this block will be distributed across the entire physical media. In calculating the time before the media wears out it actually does not matter if the same host block is written or different host blocks, the calculation is the same.

To calculate the minimum time it would take for the flash to wear out regardless of which host blocks are written is:
Time = (Physical Media Capacity * Maximum Write Cycles) / Maximum Write Speed

The disk which has 18Gb of host addressable storage actually has 32Gb of physical media (* this is unique to the drives used in SpeedStor). The physical media capacity for the smallest disk 18Gbytes, the maximum write cycles has been tested to 8 million so we can set a very conservative figure of 1 million, the maximum write speed of the drive is 100MB/s. If you plug this into the formula:

(34,359,738,368 * 1000000) / 100000000 = 343,597,383 seconds

This equates to about 11 years. Obviously if the drive is of higher capacity this time is extended dramatically (we figure 20 years for the 36GB unit). Again this is using 1 million not 8 million, which we have tested***. Also if the host pauses during these 11 years of writing to read the data then again the time is extended.

Q: Are NAND based systems a cheap second-choice systems to the DRAM based systems?

A: Although NAND SSD conjures up thoughts of low end, inexpensive SSDs, nothing could be further from the truth. There are two levels of NAND SSDs. Those made with single layer cell (SLC) chips and those made with multiple layer chips (MLC). Commercial grade drives are made with MLC’s and targeted for use in laptops or hand-held devices. Commercial grade drives also do not incorporate high-end ECC and Flash management.

The SpeedStor is the only SSD manufactured to compete in the same marketplace of DRAM based systems…our enterprise customers. To compete with a DRAM SSD the drive must be extremely reliable and it must be able to sustain high IOPs.

The NAND industry and traditional customer base would not be the same customer base as what we are referring to as “enterprise class”…that being a customer who seeks maximum IO performance, maximum reliability, and enterprise class interfaces (FC). The SpeedStor has taken the benefits of NAND and designed around its pitfalls (write performance and life of cells) thereby offering all of the benefits of a DRAM SSD but without its pitfalls (volatile, expensive, large)

Q: How does SpeedStor fit into the overall SSD Market?

Three types of NAND markets:

1. Commercial: Think laptops, handhelds etc. Very inexpensive. Low reliability, low cost, low performance (2k-7.5k IOPS, up to 45Mbs transfer rate). Use SLC (single level cell) chips and moving to MLC (multi level cell) chips (which are slower and less reliable). Primarily use for size and reliability as opposed to performance. Cost of an 18Gb drive is less than $1k. These drives implement the NAND or Nor chips with little engineering around them. You are essentially as strong as the chip.

2. Military NAND customers: High reliability, higher in cost, low performance (8,000IOPs, 50 MBs. Transfer rate). This market is generally interested in SSD due to the fact that there are no moving parts and can therefore sustain the heat and vibration that they may encounter if deployed on a vehicle (planes, tanks, etc). Specifically interested in NAND because it is non volatile and not prone to loss of data in event of power loss. (can’t use DRAM because they require a battery to disk backup system that will not sustain the vibration and heat requirements).

These drives use high- end commercial or industrial chips (designed for higher temperature fluctuations). These drives are designed for maximum reliability. Only SLC (Single layer chips) are used and generally excellent wear leveling techniques are implemented to guard against the limitation of 100k writes/cell.

Military SSDs are either designed with commercial grade design (normal heat dissipation and vibration requirements as compared to the enterprise) or with industrial grade (to meet extreme shake and heat requirements). The commercial grade systems cost roughly the same amount as the SpeedStor but are significantly slower in terms of both IOPS as well as sustained transfer rates.

3. Enterprise: Introducing SpeedStor. A NAND SSD designed for the enterprise. Performance like that of a DRAM SSD at the price of a NAND SSD. Our solutions range between $10k (single 146GB drive in 1U redundant enclosure) and $32k with four 146GB SSDs in fully redundant enclosure.

Q: Do you offer SpeedStore in an HA configuration.

A: Yes we offer a “mirrored paired” SpeedStore configuration as a high availability solution. The solution is sold on performance for low dollars not MBs/dollar. You can offer 10X performance, better usage of infrastructure, lower TCO, maximum reliability for as little as $18K.

FAQ/NPR #9: Moving my existing metadata will put my data at risk and disrupt access to the system

The SSD is installed as a mirrored pair. If the metadata is presently installed on a mirrored set of drives., simply break the mirror set, copy the offline set to the new SSD and do a shadow copy update. There is no risk of loss of data or access to the system.

Q: In the unlucky event of a SpeedStore failure, how do I get support?

A: The drives are designed such that there is no maintenance, but in the unlikely event that there is a problem, the drive will be replaced.

SDS offers all levels of SLA’s:

Q: Is it true that every application using SSD will perform better than on spinning disk?

A: True. One SpeedStor drive is comparable to the capability of 200 rotating drives in a random environment. If in a single stream sequential, large block environment, then one is limited by the transfer rate of a file from and to the server, and the drives will appear to be performing the same. Basically the drive is yawning and waiting for more to do… In order to benefit from the low latency of the SSD one must have multiple paths of IO.