New Dell FX2 chassis mixes 'n' matches server innards Dell is making its converged IT play with a new chassis that can blend compute, networking and storage kit in many configurations to allow the assembly of 2U machines tuned to different workloads. The new beast is called the PowerEdge FX2 Enclosure and Dell's building a range of modules with which you can fill it. Whatever you …
I'm using my imagination! Like from the first line I am imagining a literal blender built into a full size cabinet, where all the parts of a computer go. They are blended together and then extruded into extremely standardized lumps of computer parts, like chickens blended into McNuggets!
And that's where Optiplexes come from!
Did you even see one? Maybe I'm wrong, you might be an enthusiast who's really excited about this new kit. However when an "article" doesn't even suggest there might be a negative about a product, readers get suspicious. You could at least have said "Drawbacks? none that I've come across so far." Or "regurgitated press release, oh yeah that's how I do my job".
There is so much to this FX architecture and design that you really need to review the product details. Just to name a few below. If I am wrong please feel free to correct me regarding Twin series. The difference between Supermicro Twin series and Dell FX2 is like a cheap version of an Android Phone and an elegant design of Apple iPhone 6 (FX in this case). You have to get your hands on one of these to experience it. If you don't like it, come back and post your experience :)
1). First of all there is no IO Aggregation of networking in Twin series. A huge upside of using FX2 with 10Gb networking already switched inside the system for all cases of server sleds. A big opex and capex savings.
2). The sleds are all serviceable from the front like any rack servers. Not from the hot aisle like in case of twin series.
3). Management is truly converged for all Storage, Networking and Compute. Just give it a shot. There is no comparison.
4). A lot of PCIe slots and can be assigned depending on the workload needs to appropriate compute in case storage is used in other sleds
5). Storage density is industry leading with still keeping the full Hot swappability and High Availability
6). Very cost effective SSD density and performance is finally in servers with still keeping 2U formfactor with full serviceability and upto 8 x 2S High end XEON CPU
….List goes on …
BTW, I don't work at Dell but glad to see a very innovative and game changing product launched by a company. Getting the best of all worlds in one design- Blades, Racks and Converged Infrastructure while eliminating most of the negative aspects from those. Wait till HP, Lenovo and Supermicro types will be copying the design. But I am sure there so many patents on this by Dell that others will have to watch out!
I do work at Dell and was on the team that conceived and designed FX2. Thanks for your comments and you are correct, plus:
1) FX2 supports a much wider range of compute options than any other solution on the market
-Atom based Microservers
-Ultra high density 2S Xeon (highest density 2S server i know of on the market - FC430 coming early next year)
-Mainstream 2S Xeon with 50% more memory & IO capacity than the SuperMicro Twins or any other 4in2U type of design
-4S Xeon (FC830 coming early next year)
2) DAS storage in much higher capacity (comparing 2.5") and configurability than any other "twin" type of solution (enables scaling from 8 to 48 drives per server). We built FX2 to play well in the world of centralized SAN/NAS storage, but we optimized it around the new storage paradigms which are built around internal & external DAS (vSAN, shared nothing, scale out database, Exchange, Microsoft Windows Storage Spaces...).
Hi Mike,
FX2 supports a much wider range of compute options than any other solution on the market
It's interesting you mention this as a selling feature. Currently, I do not see GPU support for the FX2, whereas I see this available in the Supermicro Twin Family. I even have a GPU FatTwin of my own, and will be doing a review on it here for The Register shortly. All the testing is done, I just have to write it.
Atom based Microservers
While not a part of the Twin Family officially, the MicroBlade series Supermicro punts is frequently considered a "close cousin", and offers the best density for Atom-based Microservers I've seen outside of a SeaMicro setup. (28 nodes in 6U or 784 nodes in a 42U rack.)
The Supermicro 2U Twin3 gets 8 Atom nodes into 2 U.
How does FX2 compare?
Ultra high density 2S Xeon (highest density 2S server i know of on the market - FC430 coming early next year)
Please correct me if I'm wrong, but you get 4x 2P servers into 2U, correct?
Both the 2U Twin2 and the 2U Twin Pro (as well, I think the 2U Twin Pro2) have been achieving that density for some time. With dual 10Gbase-T and heft storage per node, if that's what you want, all included.
If you've crammed more Xeons into 2U with enough room for disks and RAM, hurray! I'd love to know more!
4S Xeon (FC830 coming early next year)
This is something I wish Supermicro did.
DAS storage in much higher capacity (comparing 2.5") and configurability than any other "twin" type of solution (enables scaling from 8 to 48 drives per server). We built FX2 to play well in the world of centralized SAN/NAS storage, but we optimized it around the new storage paradigms which are built around internal & external DAS (vSAN, shared nothing, scale out database, Exchange, Microsoft Windows Storage Spaces...).
Are you claiming you have put 48 drives physically into 2U? If so, I am impressed. Supermicro's TwinPro only manages to get 24 drives into 2U, albeit, that is also whilst cramming 4 compute nodes into the same space.
Even Supermicro's Storage Bridge Bay stuff (basically two nodes with a shared backplane to two-port SAS drives for roll-your-own SAS) is 24 2.5" in 2U.
The closest I can come to 48 physical drives in 2U within the Supermicro world is the 6047R-E1R72L2K which is 72 3.5" drives in 4U, or about 36 3.5" drives in 2U.
That makes me wonder, what drives size is that 48 drives? Even at 2.5", I physically don't understand how you put that many drives into 2U and still have room for a computer.
I'd love to learn more, some day. It looks like Dell has some interestingly competitive offerings with it's FX2 line. I guess price will tell...and whether the inevitable premium is justified by the management tools, software shipped and the higher build quality of items like the rail kits.
Cheers, and congrats on bringing to market something that legitimately qualifies as "nerd porn".
Hi Trevor
1) No GPUs in FX2, we'll have a GPU optimized solution coming soon. we looked at that quite a bit and once you start putting GPUs into shared infrastructure you have to reduce the compute nodes to account for the power/space needs of the gpu so you lose the advantage of shared infrastructure (density, cost...). Our take is GPUs are better done in optimized rackmounts, arguements can be made on both sides i'm sure. you'll see something coming from us in a few weeks in that vein.
2) we can get 16 Atom servers in 2U w/ FX2
3) we can get 8 x 2S Xeons (v3) in FX2. the node is called FC430, its a quarter width compute node and will come out in first quarter next year. it has limited onboard storage (2 x hot plug SSDs) but can be used in conjunction with the FD332 storage sled. so you can get 8 x FC430s with the onboard hot plug SSDs in a 2U FX2 or 4 x FC430s + 2 storage sles giving you 4 servers each with 2 x SSDs + 8 x 2.5" HDD/SSD
4) the FD332 storage sled has 16 x 2.5" drives in a drawer, drives hot plug from the side as you slide out the drawer (drives stay powered on as drawer slides out). the sled comes with 1 or 2 x PERC9 H730p 12Gb SAS RAID controllers (also run in HBA mode). with the upcoming FD332 storage sled we can get:
-1x2S server (each w/ 8 x SSDs or 2 x 2.5" HDD/SSD/NVMe) + 48 x 2.5" drives in 2U. all disks hot pluggable
-1x4S server (with 16 x SSDs or 8 x 2.5" HDD/SSD) + 32 x 2.5" drives in 2U
-2x2S servers (each with 8 x SSDs or 2 x 2.5" HDD/SSD/NVMe) + 32 x 2.5" drives (16 per server) in 2U
-4x2S servers (with 2 x SSDs) + 32 x 2.5" drives (8 per server) in 2U
The FC630 (1/2 width node 4 per 2U chassis) has option for 8 x 1.8" SSDs up to 800GB so in many cases that will be plenty of capacity and certainly storage performance for many appliations including all flash vSANs without even adding a storage sled.
The difference between Supermicro Twin series and Dell FX2 is like a cheap version of an Android Phone and an elegant design of Apple iPhone 6 (FX in this case).
That's typically the way Dell versus Supermicro works, yep.
First of all there is no IO Aggregation of networking in Twin series. A huge upside of using FX2 with 10Gb networking already switched inside the system for all cases of server sleds. A big opex and capex savings.
A huge advantage of the Twin series is that there is no networking aggregation, thus it is a "dumb" (and hence cheap chassis). My understanding is that he FX is a little bit more like someone took the blade concept but reinvented it into a Twin form factor.
The sleds are all serviceable from the front like any rack servers. Not from the hot aisle like in case of twin series.
Supermicro Twin series have both front and rear I/O for every configuration I am aware of.
Management is truly converged for all Storage, Networking and Compute. Just give it a shot. There is no comparison.
Supermicro provides just raw hardware. Management is IPMI via the BMC. That's it. It's very bare bones. I'd love to see what Dell offers in terms of management that goes beyond this, but their PR is well night impenetrable.
A lot of PCIe slots and can be assigned depending on the workload needs to appropriate compute in case storage is used in other sleds
if you take a look at the Caesium cluster in my lab, I think you'll see it is based on the F627G2-F73PT+ Fat Twin. This comes with many PCI-E slots, and in fact two of the four nodes in Caesium currently contain nVidia GRID cards.
Storage density is industry leading with still keeping the full Hot swappability and High Availability
I have to debate this claim, sir. Pleas see Supermicro's Twin family page. They have Compute, GPU, Storage Heavy and even Hadoop-specialised members of the Twin family. I strongly doubt you're going to cram more disks per node into something than Supermicro have achieved...though I'd be quite pleased to be wrong.
Very cost effective SSD density and performance is finally in servers with still keeping 2U formfactor with full serviceability and upto 8 x 2S High end XEON CPU
Again, very standard stuff with Supermicro. Been around for ages in the Twin line.
BTW, I don't work at Dell but glad to see a very innovative and game changing product launched by a company. Getting the best of all worlds in one design- Blades, Racks and Converged Infrastructure while eliminating most of the negative aspects from those. Wait till HP, Lenovo and Supermicro types will be copying the design. But I am sure there so many patents on this by Dell that others will have to watch out!
Actually, this looks very much like Dell copied the Supermicro Fat Twin, refined it, added layers of "luxury" that can drive up the margins and basically created a solution that competes with the already fourth or fifth generation Supermicro Twin line but for companies less willing to work with bare-bones, stripped down gear.
I admit this is a required and desired niche, but I will require quite a bit of convincing to believe that it qualifies as "game changing". I say this mostly because I personally am on my second generation of Fat Twin, and I see most of the benefits you've described already. That makes this iterative, not revolutionary.
I am, however, entirely happy to be proven wrong.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
