SSD performance – Page 3 – Silverton Consulting

SSD market dynamics

Posted on March 18, 2011April 10, 2012 by Ray in Market dynamics, Storage, storage economics, Strategic Inflection Points

Toshiba's 2.5" SSD (from SSD.Toshiba.com)

Had a talk the other week with an storage executive about SSD and NAND cost trends. It seemed that everyone thought that $/GB for SSD was going to overtake (be less costly) than enterprise class disk sometime in 2013. But it appeared that NAND costs weren’t coming down as fast as anticipated and now this was going to take longer than expected.

A couple of other things are going on in the enterprise disk market that are also having an effect on the relative advantage of SSDs over disks. Probably, most concerning to SSD market is enterprise storage’s new penchant for sub-LUN tiering.

Automated sub-LUN storage tiering

The major storage vendors all currently support some form of automated storage tiering for SSD storage (NetApp’s Flash Cache does this differently but the impact on NAND storage requirements is arguably similar). Presumably, such tiering should take better advantage of any amount of SSD/NAND storage available to a storage system.

Prior to automated sub-LUN storage tiering, one had to move a whole LUN to SSDs to take advantage of its speed. However, I/O requests or access are not necessarily at the same intensity for all blocks of a LUN. So one would typically end up with an SSD LUN with a relatively few blocks being heavily accessed while the vast majority of its blocks would not be being hit that much. We paid the high price of SSD LUNs gladly to get the high performance for those few blocks that really needed it.

However, with sub-LUN tiering or NAND caching, one no longer has to move all the blocks of a LUN into NAND storage to gain its benefits. One can now just have the system identify those select blocks which need high performance and move those blocks and those blocks only to NAND storage. The net impact of sub-LUN tiering or NAND caching is that one should require less overall NAND storage to obtain the same performance as one had previously with SSDs alone.

On the other hand, some would say that making the performance advantages of NAND be available at a lower overall cost might actually increase the overall amount of NAND shipments. Also with automated sub-LUN tiering in place, this removes all the complexity needed previously to identify which LUNs needed higher performance. Reducing such complexity should increase SSD or NAND market penetration.

Nonetheless, I feel that given todays price differential of SSDs over enterprise disk, the people buying SSDs today have a very defined need for speed and would have paid the price anyways for SSD storage. Anything we do to make satisfying that need with less SSD or NAND storage should reduce the amount of SSDs shipped today.

But getting back to that price crossover point, as the relative price of NAND on $/GB comes down, having an easy way to take advantage of its better performance should increase its market adoption, even faster than price would do alone.

Comments?

SCI’s latest SPECsfs2008 NFS ops vs. system size – chart of the month

Posted on January 11, 2011August 19, 2016 by Ray in File Storage, NFS throughput, SPECsfs, SPECsfs2008, Storage, Storage architecture, Storage performance

(c) 2011 Silverton Consulting, Inc., All Rights Reserved

We return to our periodic discussion of storage system performance, this time on SPECsfs(r)2008 benchmark results for NFS file serving ops per second vs. system reported memory size. For some obscure reason, I was very intrigued with this chart.

Chart description

We have broken the data out to show those system that only used DRAM in system memory with only hard disk drives and those systems that also included either NAND cache in system memory or SSDs. Current SPECsfs2008 results show 33 systems with DRAM and disk drives and only 6 using SSDs or NAND cache for NFS results.

The horizontal axis shows system memory size for the systems under test and doesn’t include SSD capacity (considered drive capacity by SPECsfs2008) size but does include NAND cache size (considered system memory by SPECsfs2008). The vertical axis shows maximum NFS throughput operations per second attained by the storage. The two lines are Excel generated linear regressions across the two sets of data (DRAM-Disk only systems and SSD or NAND caching systems).

Discussion of results

Given the limited data we probably can’t conclude much from the SSD-NAND linear regression line other than it’s different and somewhat less than what can be gained on average from a system using DRAM and disk only. Also the regression coefficient (R**2) of either linear regression is not that great (~0.62 for DRAM-Disk only and ~0.69 for SSD or NAND use) which might be stretching any real discussion based on statistical normalcy. Nevertheless, one question that emerges is why would SSD or NAND use not generate an relatively equivalent amount of NFS throughput as systems with DRAM only?

Obviously, NAND-SSDs access times are not as fast as DRAM. Thus, if I had 800GB of DRAM, I could potentially access data faster than if I had 800GB of NAND cache or SSDs, all things being equal. However, when talking about access time frames at the sub-msec level one would think it wouldn’t make that much of a difference, but apparently it does.

Also, given the limited data on SSDs vs. NAND cache use, it’s hard to make any distinction between these two types of systems but one can hope that as more data comes in, we can answer this as well. Another question is whether SSD capacity should be considered system memory or drive capacity for benchmark purposes. SPECsfs2008 states that SSD size should be considered drive capacity but that’s open to some debate which more data could help resolve.

Finally, missing from SPECsfs2008 reports is any statement of system cost. As such, it’s impossible to add any measure of relative cost effectiveness factor to this discussion. However, given the current price differential (on $/GB) between DRAM and NAND memory or SSD capacity, one could probably conclude that on a cost effectiveness basis the relative advantages of DRAM only systems might diminish. But without system cost information that’s difficult to quantify.

—-

The full performance dispatch will be up on our website after month end but if one is interested in seeing it sooner sign up for our free monthly newsletter (see subscription widget, above right) or subscribe by email and we will send the current issue along with download instructions for this and other reports. If you need an even more in-depth analysis of NAS system performance please consider purchasing SCI’s NAS Buying Guide also available from our website.

As always, we welcome any constructive suggestions on how to improve any of our storage performance analysis.

Comments?

Top 10 storage technologies over the last decade

Posted on December 30, 2010April 10, 2012 by Ray in Block Storage, Cloud services, Data, Data compression, Disk storage, Ethernet, FC, File Storage, Networking, Storage, Storage Backup, Storage density, Storage Features, Storage performance, Strategic Inflection Points, System effectiveness, Systems, Tape storage

Aurora's Perception or I Schrive When I See Technology by Wonderlane (cc) (from Flickr)

Some of these technologies were in development prior to 2000, some were available in other domains but not in storage, and some were in a few subsystems but had yet to become popular as they are today. In no particular order here are my top 10 storage technologies for the decade:

NAND based SSDs – DRAM and other technology solid state drives (SSDs) were available last century but over the last decade NAND Flash based devices have dominated SSD technology and have altered the storage industry forever more. Today, it’s nigh impossible to find enterprise class storage that doesn’t support NAND SSDs.
GMR head– Giant Magneto Resistance disk heads have become common place over the last decade and have allowed disk drive manufacturers to double data density every 18-24 months. Now GMR heads are starting to transition over to tape storage and will enable that technology to increase data density dramatically
Data Deduplication – Deduplication technologies emerged over the last decade as a complement to higher density disk drives as a means to more efficiently backup data. Deduplication technology can be found in many different forms today, ranging from file and block storage systems, backup storage systems, to backup software only solutions.
Thin provisioning – No one would argue that thin provisioning emerged last century but it took the last decade to really find its place in the storage pantheon. One almost cannot find a data center class storage device that does not support thin provisioning today.
Scale-out storage – Last century if you wanted to get higher IOPS from a storage subsystem you could add cache or disk drives but at some point you hit a subsystem performance wall. With scale-out storage, one can now add more processing elements to a storage system cluster without having to replace the controller to obtain more IO processing power. The link reference talks about the use of commodity hardware to provide added performance but scale-out storage can also be done with non-commodity hardware (see Hitachi’s VSP vs. VMAX).
Storage virtualization – server virtualization has taken off as the dominant data center paradigm over the last decade but a counterpart to this in storage has also become more viable as well. Storage virtualization was originally used to migrate data from old subsystems to new storage but today can be used to manage and migrate data over PBs of physical storage dynamically optimizing data placement for cost and/or performance.
LTO tape – When IBM dominated IT in the mid to late last century, the tape format dejour always matched IBM’s tape technology. As the decade dawned, IBM was no longer the dominant player and tape technology was starting to diverge into a babble of differing formats. As a result, IBM, Quantum, and HP put their technology together and created a standard tape format, called LTO, which has become the new dominant tape format for the data center.
Cloud storage – Unclear just when over the last decade cloud storage emerged but it seemed to be a supplement to cloud computing that also appeared this past decade. Storage service providers had existed earlier but due to bandwidth limitations and storage costs didn’t survive the dotcom bubble. But over this past decade both bandwidth and storage costs have come down considerably and cloud storage has now become a viable technological solution to many data center issues.
iSCSI – SCSI has taken on many forms over the last couple of decades but iSCSI has the altered the dominant block storage paradigm from a single, pure FC based SAN to a plurality of technologies. Nowadays, SMB shops can have block storage without the cost and complexity of FC SANs over the LAN networking technology they already use.
FCoE – One could argue that this technology is still maturing today but once again SCSI has taken opened up another way to access storage. FCoE has the potential to offer all the robustness and performance of FC SANs over data center Ethernet hardware simplifying and unifying data center networking onto one technology.

No doubt others would differ on their top 10 storage technologies over the last decade but I strived to find technologies that significantly changed data storage that existed in 2000 vs. today. These 10 seemed to me to fit the bill better than most.

Comments?

SCI’s latest SPC-1&-1/E LRT results – chart of the month

Posted on December 20, 2010October 8, 2012 by Ray in Block Storage, LRT, SPC-1, Storage, Storage performance, Storage Performance Council, System effectiveness, Systems

(c) 2010 Silverton Consulting, Inc., All Rights Reserved

It’s been a while since we reported on Storage Performance Council (SPC) Least Response Time (LRT) results (see Chart of the month: SPC LRT[TM]). This is one of the charts we produce for our monthly dispatch on storage performance (quarterly report on SPC results).

Since our last blog post on this subject there have been 6 new entries in LRT Top 10 (#3-6 &, 9-10). As can be seen here which combines SPC-1 and 1/E results, response times vary considerably. 7 of these top 10 LRT results come from subsystems which either have all SSDs (#1-4, 7 & 9) or have a large NAND cache (#5). The newest members on this chart were the NetApp 3270A and the Xiotech Emprise 5000-300GB disk drives which were published recently.

The NetApp FAS3270A, a mid-range subsystem with 1TB of NAND cache (512MB in each controller) seemed to do pretty well here with all SSD systems doing better than it and a pair of all SSD systems doing worse than it. Coming in under 1msec LRT is no small feat. We are certain the NAND cache helped NetApp achieve their superior responsiveness.

What the Xiotech Emprise 5000-300GB storage subsystem is doing here is another question. They have always done well on an IOPs/drive basis (see SPC-1&-1/E results IOPs/Drive – chart of the month) but being top ten in LRT had not been their forte, previously. How one coaxes a 1.47 msec LRT out of a 20 drive system that costs only ~$41K, 12X lower than the median price(~$509K) of the other subsystems here is a mystery. Of course, they were using RAID 1 but so were half of the subsystems on this chart.

It’s nice that some turnover in this top 10 LRT. I still contend that response time is an important performance metric for many storage workloads (see my IO throughput vs. response time and why it matters post) and improvement over time validates my thesis. Also I received many comments discussing the merits of database latencies for ESRP v3 (Exchange 2010) results, (see my Microsoft Exchange Perfomance ESRP v3.0 results – chart of the month post). You can judge the results of that lengthy discussion for yourselves.

The full performance dispatch will be up on our website in a couple of weeks but if you are interested in seeing it sooner just sign up for our free monthly newsletter (see upper right) or subscribe by email and we will send you the current issue with download instructions for this and other reports.

As always, we welcome any constructive suggestions on how to improve our storage performance analysis.

Comments?

The future of data storage is MRAM

Posted on November 30, 2010April 12, 2012 by Ray in Storage, Storage density, Storage longevity, Storage performance, System effectiveness, Systems

We have been discussing NAND technology for quite awhile now but this month I ran across an article in IEEE Spectrum titled “a SPIN to REMEMBER – Spintronic memories to revolutionize data storage“. The article discussed a form of magneto-resistive random access memory or MRAM that uses quantum mechanical spin effects or spintronics to record data. We have talked about MRAM technology before and progress has been made since then.

Many in the industry will recall that current GMR (Giant Magneto-resistance) heads and TMR (Tunnel magneto-resistance) next generation disk read heads already make use of spintronics to detect magnetized bit values in disk media. GMR heads detect bit values on media by changing its electrical resistance.

Spintronics however can also be used to record data as well as read it. These capabilities are being exploited in MRAM technology which uses a ferro-magnetic material to record data in magnetic spin alignment – spin UP, means 0; spin down, means 1 (or vice versa).

The technologists claim that when MRAM reaches its full potential it could conceivably replace DRAM, SRAM, NAND, and hard disk drives or all current electrical and magnetic data storage. Some of MRAM’s advantages include unlimited write passes, fast reads and writes and data non-volatilility.

MRAM reminds me of old fashioned magnetic core memory (in photo above) which used magnetic polarity to record non-volatile data bits. Core was a memory mainstay in the early years of computing before the advent of semi-conductor devices like DRAM.

Back to future – MRAM

However, the problems with MRAM today are that it is low-density, takes lots of power and is very expensive. But technologists are working on all these problems with the view that the future of data storage will be MRAM. In fact, researchers at the North Carolina State University (NCSU) Electrical Engineering department have been having some success with reducing power requirements and increasing density.

As for data density NCSU researchers now believe they can record data in cells approximating 20 nm across, better than current bit patterned media which is the next generation disk recording media. However reading data out of such a small cell will prove to be difficult and may require a separate read head on top of each cell. The fact that all of this is created with normal silicon fabrication methods make doing so at least feasible but the added chip costs may be hard to justify.

Regarding high power, their most recent design records data by electronically controlling the magnetism of a cell. They are using dilute magnetic semiconductor material doped with gallium maganese which can hold spin value alignment (see the article for more information). They are also using a semiconductor p-n junction on top of the MRAM cell. Apparently at the p-n junction they can control the magnetization of the MRAM cells by applying -5 volts or removing this. Today the magnetization is temporary but they are also working on solutions for this as well.

NCSU researchers would be the first to admit that none of this is ready for prime time and they have yet to demonstrate in the lab a MRAM memory device with 20nm cells, but the feeling is it’s all just a matter of time and lot’s of research.

Fortunately, NCSU has lots of help. It seems Freescale, Honeywell, IBM, Toshiba and Micron are also looking into MRAM technology and its applications.

—–

Let’s see, using electron spin alignment in a magnetic medium to record data bits, needs a read head to read out the spin values – couldn’t something like this be used in some sort of next generation disk drive that uses the ferromagnetic material as a recording medium. Hey, aren’t disks already using a ferromagnetic material for recording media? Could MRAM be fabricated/layed down as a form of magnetic disk media?? Maybe there’s life in disks yet….

What do you think?

SNIA’s new SSD performance test specification

Posted on July 13, 2010April 10, 2012 by Ray in Storage, Storage performance

Western Digital's Silicon Edge Blue SSD SATA drive (from their website)

A couple of weeks ago SNIA ~~just~~ released a new version of their SSSI (SSD) performance test specification for public comment. Not sure if this is the first version out for public comment or not but I discussed a prior version in a presentation I did for SNW last October and I have blogged before about some of the mystery of measuring SSD performance. The current version looks a lot more polished than what I had to deal with last year but the essence of the performance testing remains the same:

Purge test – using vendor approved process, purge (erase) all the data on the drive.
Preconditioning test – Write 2X the capacity of the drive using 128KiB blocksizes and sequentially writing through the whole device’s usable address space.
Steady state testing – varying blocksizes, varying read-write ratios, varying block number ranges, looped until steady state is achieved in device performance.

The steady state testing runs a random I/O mix for a minutes duration at whatever the current specified blocksize, RW ratio and block number range. Also, according to the specification the measurements for steady state are done once 4KiB block sizes and 100% ~~Read~~ Write settles down. This steady state determinant testing must execute over a number of rounds (4?) then the other performance test runs are considered at “steady state”.

SNIA’s SSSI performance test benefits

Lets start by saying no performance test is perfect. I can always find fault in any performance test, even my own. Nevertheless, the SSSI new performance test goes a long way towards fixing some intrinsic problems with SSD performance measurement. Specifically,

The need to discriminate between fresh out of the box (FOB) performance and ongoing drive performance. The preconditioning test is obviously a compromise in attempting to do this but writing double the full capacity of a drive will take a long time and should cause every NAND cell in the user space to be overwritten. Once is not enough to overwrite all the devices write buffers. However three times the device’s capacity may still show some variance in performance but it will take correspondingly longer.
The need to show steady state SSD performance versus some peak value. SSDs are notorious for showing differing performance over time. Partially this is due to FOB performance (see above) but mostly this is due to the complexity of managing NAND erasure and programming overhead.

The steady state performance problem is not nearly as much an issue with hard disk drives but even here, with defect skipping, drive performance will degrade over time (but a much longer time than for SSDs). My main quibble with the test specification is how they elect to determine steady state – 4KiB with 100% ~~read~~ write seems a bit over simplified.

Is write some proportion of read IO needed to define SSD “steady state” performance?

[Most of the original version of this post centered on the need for some write component in steady state determination. This was all due to my misreading the SNIA spec. I now realize that the current spec calls for a 100% WRITE workload with 4KiB blocksizes to settle down to determine steady state. While this may be overkill, it certainly is consistent with my original feelings that some proportion of write activity needs to be a prime determinant of SSD steady state.]

~~Most of my concern with how the test determines SSD steady state performance is that lack of write activity.~~ One concern is the lack of read activity in determining steady state. My other worry with this approach is the blocksize seems a bit too small, however this is minor in comparison.

Let’s start with the fact that SSDs are by nature assymetrical devices. By that I mean their write performance differs substantially from their read performance due to the underlying nature of the NAND technology. But much of what distinguishes an enterprise SSD from a commercial drive is the sophistication of its write processing. ~~By using a 100% read rate we are undervaluing this sophistication.~~

But using 100% writes to test for steady state may be too much.

In addition, it’s It is hard for me to imagine any commercial or enterprise class device in service not having some high portion of ongoing ~~write~~ read IO activity. I can easily be convinced that a normal R:W activity for an SSD device is somewhere between 90:10 and 50:50. But I have a difficult time seeing an SSD R:W ratio of ~~100:0~~ 0:100 as realistic. And I feel any viable interpretation of device steady state performance needs to be based on realistic workloads.

In SNIA’s defense they had to pick some reproducible way to measure steady state. Some devices may have had difficulty reaching steady state with ~~any~~ 100% write activity. However, most other benchmarks have some sort of cut off that can be used to invalidate results. Reaching steady state is one current criteria for SNIA’s SSSI performance test. I just think adding some ~~portion of write~~ mix of read and write activity would be a better measure of SSD stability.

As for the 4KiB block size, it’s purely a question of what’s the most probable blocksize in the use of SSDs and may vary for enterprise or consumer applications. But 4KiB seems a bit behind the times especially with todays 128GB and higher drives…

What do you think should SSD steady state need some ~~portion of write~~ mix of read and write activity or not?

[Thanks to Eden Kim and his team at SSSI for pointing out my spec reading error.]

SPC-1 Results IOPs vs. Capacity – chart of the month

Posted on March 23, 2010October 8, 2012 by Ray in IOPS, SPC-1, Storage, Storage performance, Storage Performance Council

SPC-1* IOPS vs. Capacity, (c) 2010 Silverton Consuliting, All Rights Reserved

This chart is from SCI’s last months report on recent Storage Performance Council (SPC) benchmark results. There were a couple of new entries this quarter but we decided to introduce this new chart as well.

This is a bubble scatter plot of SPC-1(TM) (online transaction workloads) results. Only storage subsystems that cost less than $100/GB, trying to introduce some fairness.

Bubble size is a function of the total cost of the subsystem
Horizontal access is subsystem capacity in GB
Vertical access is peak SPC-1 IOPS(TM)

Also we decided to show a linear regression line and equation to better analyze the data. As shown in the chart there is a pretty good correlation between capacity and IOPS (R**2 of ~0.8). The equation parameters can be read from the chart but it seems pretty tight from a visual perspective.

The one significant outlier here at ~250K IOPS is TMS RAMSAN which uses SSD technology. The two large bubbles at the top right were two IBM SVC 5.1 runs at similar backend capacity. The top SVC run had 6 nodes and the bottom SVC run only had 4.

As always, a number of caveats to this:

Not all subsystems on the market today are benchmarked with SPC-1
The pricing cap eliminated high priced storage from this analysis
IOPS may or may not be similar to your workloads.

Nevertheless, most storage professionals come to realize that having more disks can often result in better performance. This is often confounded by RAID type used, disk drive performance, and cache size. However, the nice thing about SPC-1 runs, is that most (nearly all) use RAID 1, have the largest cache size that makes sense, and the best performing disk drives (or SSDs). The conclusion cannot be more certain – the more RAID 1 capacity one has the higher the number of IOPS one can attain from a given subsystem.

The full SPC report went out to our newsletter subscribers last month and a copy of the report will be up on the dispatches page of our website later this month. However, you can get this information now and subscribe to future newsletters to receive future full reports even earlier, just email us at SubscribeNews@SilvertonConsulting.com?Subject=Subscribe_to_Newsletter.

As always, we welcome any suggestions on how to improve our analysis of SPC or any of our other storage system performance results. This new chart was a result of one such suggestion.

WD’s new SiliconEdge Blue SSD data write spec

Posted on March 4, 2010April 10, 2012 by Ray in Storage, Storage reliability

Western Digital (WD) announced their first SSD drive for the desktop/laptop market space today. Their drive offers the typical256, 128, and 64GB capacity points over a SATA interface. Performance looks ok at 5K random read or write IO/s with sustained transfers at 250 and 140MB/s for read and write respectively. But what caught my eye was a new specification I hadn’t seen before indicating Maximum GB written per day of 17.5, 35 and 70GB/d for their drives using WD’s Operational Lifespan – LifeEST(tm) definition.

I couldn’t find anywhere that said which NAND technology was used in the device but it likely uses MLC NAND. In a prior posting we discussed a Toshiba study that said a “typical” laptop user writes about 2.4GB/d and a “heavy” laptop user writes about 9.2GB/d. This data would indicate that WD’s new 64GB drive can handle almost 2X the defined “heavy” user workload for laptops and their other drives would handle it just fine. A data write rate for desktop work, as far as I can tell, has not been published, but presumably it would be greater than laptop users.

From my perspective more information on the drives underlying NAND technology, on what a LifeEST specification actually means, and a specification as to how much NAND storage was actually present would be nice, but these are all personal nits. All that aside, I applaud WD for standing up and saying what data write rate their drives can support. This needs to be a standard part of any SSD specification sheet and I look forward to seeing more information like this coming from other vendors as well.