Quick as a flash: NVMe will have you in tiers Non-Volatile Memory Express, or NVMe, is a game-changing storage standard for PCIe-connected drives. It is replacing AHCI and along with the U.2 (SFF-8639) connector it is replacing both SAS and SATA for high speed, low latency storage. It's the smart way to connect up flash and post-flash storage tech to your servers. The …
But I get the impression that for most of us the problem will be in the Network and the connectivity to these storage monsters.
As evidenced by your 'Holy Crap (batman)' when putting in a 10GbE network.
I shudder to think what sort of network connections we will need to even tickle the fancy of the next gen XFlash drives? Light Speed anyone?
The costs of going beyond 10GbE is out of the reach of many SME's so they will have to rely on Hosting or Cloud supplers for this for of setup (IMHO).
Personally, I waiting for affordable PCI-E drives of at least 1Tb capacity. One of my VM setups is making a 6core i7 4.0Ghz sweat a bit yet when I run parts of it on my 2015 Macbook pro that has PCI-E storage, it flies along.
Can you review the costs etc of moving from a smallish 1Gb Lan to 10 or beyond? This could prove very informative for those reading this article.
Well, you're looking at around £250 for a 10GbE pic-e card, and around £600 for a 8-port switch, I assume you could spend a lot more.
That said, it would make sense to only use 10GbE between servers that need it, so if it was just the matter of connecting a VM host to a SAN (for example), I'd be tempted to skip the switch and just connect directly.
Your other option would be to combine multiple 1GbE ports, but that's OS/driver/switch dependant.
40GbE has been available for a while(in 1-2U form factor switches).
The people that can't afford to buy 10GbE very very likely have no need for such storage to be connected to the network. Regular old SAS SSD platforms driving gigabytes a second(some cases 10s of gigabytes/second) and million+ IOPS are enough for just about anything.
The 500 or so VMs connected to my org's flash array combined typically pull under 100 megabytes a second on average for throughput. Burst to maybe 4-500MB/sec. Went all flash because it was cheap enough. System is connected by 6x8Gbps fibre channel connections to about 20 HP DL38x servers.
I was talking to a friend recently who is the CEO of a NVMe startup from Israel. Their tech sounds pretty good but the use cases are pretty limited. Until the costs come down enough and the standards get through enough that it just becomes a "natural progression" to use NVMe instead of SAS for SSDs, because most workloads won't see any benefit.
Most of my own workloads have seen no benefit even from going all flash, our workloads are 90%+ write, which means cache hits 99% of the time, which means low latency 99% of the time(under 2 milliseconds on average for writes) even on spinning disk because the I/O is going to the cache. For me anyway sub 5ms response times for storage are excellent, I have nothing that benefits from sub millisecond, sub 5ms becomes a rounding error in my experience anyway.
Flash gives us the capacity of course to scale up more, there's only so much data cache in the controllers(though new flash array has 5.3x more cache than previous spinning disk array). I also have to be much less concerned about unusual I/O patterns with the flash since it can take it. Though in general our I/O profiles are very predictable.
"Can you review the costs etc of moving from a smallish 1Gb Lan to 10 or beyond? This could prove very informative for those reading this article."
Done and done. Article is actually just getting tidied up now. I have a few articles on various networky bits in the hopper for this month. Stay tuned! :)
What does anyone have to say about Coho storage? "Don't fsck with the fish!"
Seriously, though. As a theoretical exercise, I like Coho. They have a good product with some great ideas for hot to expand upon its use. I also have a soft spot because they're from V-Town, and anyone who can stomach living in that overpopulated pit of parking-free madness gets my respect.
But the truth is that I haven't tested it. I've seen it. I've gotten the whole spiel on how it works, their evil plans for the future, even been shown Coho doing some very impressive things. But I haven't run my workloads on it. I haven't thrown it in my lab and tried to make it cry. I haven't done to it the horrors that I visit upon the other storage that crosses my path.
If it does what they say, as good as they say, then it's fantastic scale out storage. But I don't know where the edges are. What its tics and mannerisms and variegated vicissitudes are. I don't know how to make it stumble, how to make it fall.
Until I know that, I can't really know what I think about the product. Only what I think I probably would think about the product.
There are dozens of tech journalists who exist to republish press releases or have theoretical discussions about something they've never touched. I exist to make every piece of technology I touch cry, even to die. Then to tell you where it will work and where it won't.
That being said, they do get positive feedback from customers. So it works really well for some folks, at least. That's a great sign. :)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
