Tom's IT Pro: Mangstor MX6300 Enterprise NVMe SSD Review
NVMe provides an exceptionally high performance and low latency connection, but with a crowded field of PCIe SSD contenders, including the fabs with their price advantage, a newcomer to this space needs a differentiator. The Mangstor MX6300 is an emerging PCIe SSD that leverages the NVMe interface and offers a beefy 2.7TB of capacity, and it has a few other tricks up its sleeve.
Mangstor brings a bleeding-edge Everspin ST-MRAM buffer into the picture to protect against power failures. ST-MRAM is a persistent memory, which means it doesn't lose data when power is removed. It also offers DRAM-like speed and nanosecond latencies. ST-MRAM is one of the future memory technologies projected to replace NAND, and it's encouraging to see it already working its way into SSDs, albeit in a limited role.
The Mangstor MX6300 blends the ST-MRAM buffer in with 4GB of DRAM and an array of 3 FPGA's, all controlled by a 100-core processor running at 500MHZ. This array of components provides the fastest 4K random read/write speeds we have ever recorded from a single SSD.
The speediest performance doesn't always equate to an easy win, other considerations such as price and compatibility are also important. However, sporting a top speed of 900,000 4K random read IOPS and up to 300,000 sustained random write IOPS, the Mangstor MX6300 stakes its claim to the fastest SSD on the market. Let's see if those claims pan out as we put the MX6300 through the Tom's IT Pro wringer.
Paul Alcorn is a Contributing Editor for Tom's IT Pro, covering Storage. Follow him on Twitter and on Google+.
Follow Tom's IT Pro on Twitter, Facebook, LinkedIn and Google+.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Paul Alcorn is the Managing Editor: News and Emerging Tech for Tom's Hardware US. He also writes news and reviews on CPUs, storage, and enterprise hardware.
-
razor512 Seems the company made a mistake on the pricing, it has 1 digit too many. A more fitting price would be $1485.00Reply
No one will spend $14K on that -
palladin9479 16061529 said:Seems the company made a mistake on the pricing, it has 1 digit too many. A more fitting price would be $1485.00
No one will spend $14K on that
This isn't a consumer product, it's not even for small / medium business's. What makes it so damn expensive is the MRAM it's using as a buffer. MRAM is persistent so if there is a sudden power loss the system maintains it's state and doesn't risk data corruption. Further it's built around ridiculously high I/O rates, the kinds you only see in extremely busy database's and cloud based applications. Once you hit that level, 14K USD is absolutely nothing, support contracts and licensing costs far exceed that. For example I'll present the costs for a single Oracle Enterprise Database 11g on a single socket eight core Intel Xeon CPU.
Licensed cores = Number of Cores * Number of Sockets * Core factor. The core factor for an Intel Xeon is 0.5 per core for a total Licensed cores of 4.
http://www.oracle.com/us/corporate/pricing/technology-price-list-070617.pdf
Per Core License cost = 47,500 (per core)
Total Core license cost = $190,000 USD
Software Update License and Support = $10,450 (flat price)
Total Software License Cost = $200,450
That physical server would cost maybe $10~15K on it's own, but then you'd need an OS for it. Ms Windows is one way, the license is cheap enough same with RHEL and while technically Solaris is free, the support contract for it (required to download updates) will run you about $10K a year after the warranty period expires. When we're talking these kinds of numbers, suddenly that 14K for what is essentially super fast accelerated local storage seems kind of small and irrelevant.
-
MonsterCookie This is exactly the reason why companies who made big-iron mainframes (like SGI, SUN, Cray, IBM etc.) went bankrupt:Reply
-they thought, that other hardware/software cannot match the performance they provide, and people will keep on paying insane prices for their stuff.
-they were WRONG, and such companies will always fail, including giant software companies like Oracle.
Free software and commodity hardware might be slower, but for 200,000USD one can keep a dozen Linux servers running some other database, and pay the yearly salary of a good sysadmin.
I hope that soon that happy yellow elephant pushing that blue logo (some will know what I mean) will push hustler software companies like above out of their business. -
pedro_mann This is exactly the reason why companies who made big-iron mainframes (like SGI, SUN, Cray, IBM etc.) went bankrupt:
-they thought, that other hardware/software cannot match the performance they provide, and people will keep on paying insane prices for their stuff.
-they were WRONG, and such companies will always fail, including giant software companies like Oracle.
Free software and commodity hardware might be slower, but for 200,000USD one can keep a dozen Linux servers running some other database, and pay the yearly salary of a good sysadmin.
I hope that soon that happy yellow elephant pushing that blue logo (some will know what I mean) will push hustler software companies like above out of their business.This is exactly the reason why companies who made big-iron mainframes (like SGI, SUN, Cray, IBM etc.) went bankrupt:
-they thought, that other hardware/software cannot match the performance they provide, and people will keep on paying insane prices for their stuff.
-they were WRONG, and such companies will always fail, including giant software companies like Oracle.
Free software and commodity hardware might be slower, but for 200,000USD one can keep a dozen Linux servers running some other database, and pay the yearly salary of a good sysadmin.
I hope that soon that happy yellow elephant pushing that blue logo (some will know what I mean) will push hustler software companies like above out of their business.
Clearly, you haven't researched how expensive it is to provide a database with 900K IOPS. palladin9479 did a good job covering licensing. Hardware-wise, things get expensive quick. Hooking up a cheap cluster of linux boxes isn't going to do it. Networking alone we are talking about running 10/40GB ethernet. Just a decent mellanox switch is going to run ballpark 9K for 24 ports. If you waste most of those ports with cheap/slow servers you won't even break even on the hardware (switch + $1000-$2000 each for cheap servers.) Then you have to add in power and cooling budget for all of the "cheap" linux servers, and you run into ongoing high cost of operation.
The customers that require high performance networks know how much money this will save them in the short and long term. The future is in hyper-convergence (throwing storage and VM compute in one box and network through fat pipes) With Server 2016 density will skyrocket because a fault tolerant cluster can be built with exactly 4 servers and no shared storage (SAN etc.) costs. We used to spend 500K on a good SAN BTW (good is relative, this one card is faster), so throwing 4 of these in a tiered storage cluster is going to look cheap.
Lastly, I'de have to offer an alternate opinion of what causes big companies to die. In my mind, it is failure to perceive and embrace the future. Willingness to sacrifice their own cash cows in order to start a new billion dollar revenue stream. The reason why startups like Mangstor are going to put companies like EMC and Oracle out of business, is they don't have legacy turf to protect. And Mangstor's intended customer is able to see the savings in a gem like this. (MX6300) -
palladin9479 16064523 said:This is exactly the reason why companies who made big-iron mainframes (like SGI, SUN, Cray, IBM etc.) went bankrupt:
-they thought, that other hardware/software cannot match the performance they provide, and people will keep on paying insane prices for their stuff.
-they were WRONG, and such companies will always fail, including giant software companies like Oracle.
Free software and commodity hardware might be slower, but for 200,000USD one can keep a dozen Linux servers running some other database, and pay the yearly salary of a good sysadmin.
I hope that soon that happy yellow elephant pushing that blue logo (some will know what I mean) will push hustler software companies like above out of their business.
Don't take this as an insult, but your an amateur.
That $200,000 wasn't for a server but for the licensing costs for a single Oracle Enterprise Database 11g system. You can get it a bit cheaper if you go through one of their supply partners and buy it as part of a whole system integration piece. So purchase the RDMBS + Fusion Middle Ware + Servers + Infrastructure, whole thing will probably cost you a million or two but if your into this business then that cost doesn't scare you because it's revenue generating and amortized over the budgeted lifetime of the system.
Oracle happens to be making money hand over fist, obscene quantities of it, precisely because they tailor their products towards big data while offering full support. That last is important because when your talking hundreds of millions of USD or more worth of sales, you do not leave your companies very existence into the hands of some random Linux servers thrown together on the cheap being run by a few folks with no documentation or standardization.
RDBMS's are extremely expensive. Postgres is ok for small stuff, but it doesn't scale nor offer the same enterprise features. MS SQL is "ok" but again your lacking when it comes time to scale up. This leaves three big database's you can chose from, Oracle RDBMS, IBM DB2 and Sybase ASE. All three have their own strengths and weakness's, though only Oracle and IBM offer whole solutions. I believe Yahoo had their own software developers take Postgres apart and build a special closed source version of it for them that enabled them to scale it up. That solution isn't viable for most big customers.
This all brings me to a very important, and often overlooked point. When it comes to IT costs, hardware and licensing is dwarfed by Operation and Maintenance (O&M) costs. O&M's big ticket item is the personal costs, salary, health insurance and retirement for those IT workers that are maintaining and running the servers, possible even designing and engineering them if a company runs in-house. I've done enterprise Linux and it's no where close to what MS Windows and Solaris are for management. It requires a lot more "hands on" work though Red Hat is slowly getting there as they now offer lots of property management tools for automating large server installs. Linux is amazing as an appliance, where another vender builds the device, sells you the device and then provides support for the device, like security patch's and what not. Ultimately these are the reasons you see big data installs go with licensed software vs "free !!!" software, it is a lot cheaper in the long run and you have another entity that is legally liable for any problems or failures of their product. That all being said, Red Hat is really starting to kick a$$ right now in offering custom solutions. We are working with them to find a drop in replacement for our Solaris 10 systems running Weblogic, Oracle RDBMS and some other extremely special software. It's looking like we can get RHEL + JBOSS but are still stuck using the Oracle RDBMS's because Postgres just isn't enough for us. That should end up saving a ton of money just from switching from SPARC's to x86 in our next iteration, though it has a massive up front cost investing as we need to re-engineer our architecture from Solaris containers to VMWare eSXi (another piece of expensive software).
The close to this is that enterprise level IT functions dramatically different then medium business and lower IT. There is a line you cross that requires thinking to shift from looking at tree's to seeing the whole damn forest. You stop thinking in terms of saving a few hundred or a few thousand USD and start focusing your energy towards identifying potential problems before they become problems, then engineering solutions for those problems that can scale without inuring major cost penalties. All this requires taking an approach focused on reducing manpower requirements for doing O&M and do "more with less". -
wtfxxxgp Some of these comments were worth more to me than the article revealing this gem of a storage device O.oReply