Meet Coraid. This ten year-old startup has got its first venture capital funding and a new CEO who wants us all to understand that its AoE storage protocol is just terrific. Its product stores data in Ethernet-connected arrays using AoE - ATA over Ethernet - a block storage protocol, which, unlike Fibre Channel over Ethernet ( …
What is this guy on ?
This product sounds exactly like Adaptec's EtherScsi product developed in the late 1990's which was one of the things Adaptec brought to the effort to create iSCSI.
OK, so this product shifts T.13 ATA commands over Ethernet as opposed to T.10 SCSI commands but otherwise sounds identical - and hence, with identical issues.
Both systems work point to point based on MAC address, for a start this means that in general terms the traffic is non-routable i.e. it has to be on the same Ethernet segment. Yes, you can get a MAC level bridge, but what's the point ? Another thing not generally realised about this type of solution is that going direct over Ethernet means you're being very network unfriendly. Raw Ethernet frames will selfishly flood the network to the exclusion of all other traffic. You may achieve a decent transfer rate to the remote device, but if you're sharing the network segment with any other traffic such as SMB network shares, you're going to find the network shares dropping out at frequent intervals.
More than 10 years ago Adaptec concluded that this was an evolutionary dead-end with limited applicability. The guys who designed it have spread across senior roles in many more adventurous storage companies, funny they never proposed resurrecting the concept.
The fractional performance gain by eliminating a network friendly protocol is really not worth it and I bet the remote disks get saturated long before that even becomes a vague consideration.
USB 3.0 will kill this.
It doesn't matter
... that it's "network unfriendly". Like iSCSI, any serious deployment will run over dedicated switches or VLANs, so it can't impact front-end network operations. Likewise it's a SAN platform, so you're not going to need to route it. You'll "route" it just like you route FC, with NAS and replication hardware at the SAN edge.
HyperSCSI & EtherSCSI were crap at the time because they couldn't handle dropped frames. Once you threw more data through it than the switch could handle you started getting corrupted data. FCoE has built-in error correction.
USB 3 won't "kill this". We're talking about data centre hardware, not the quickest way to connect your porn stash to your l33t gaming rig. iSCSI, FC and FCoE are all many-to-many technologies, where multiple disk shelves can be connected to multiple servers. Sure a directly connected USB 3 hard disk may be faster, but then so is an internal SAS disk, but that doesn't matter.
SAN and LAN shouldn't mingle.
I'm not familiar with the EtherSCSI product so I won't address your comparison there, but I do think you made some interesting points about a layer2 storage protocol in general terms that should be discussed as I am familiar with the CORAID products.
AoE is indeed a layer 2 protocol which does means it is non-routable. Thus, communications are by definition contained within a network segment or VLAN. Typically, this network segment is called a SAN (Storage Area Network). In my experience, general best practices for a SAN would not include co-mingling traffic with the LAN (Local Area Network) so if one was concerned about a Storage protocol affecting the bandwidth for SMB traffic, I'd question your architecture.
AoE should be compared to typical storage protocols such as iSCSI or Fiber Channel. Rather than requiring the overhead of TCP\IP as a transport (like iSCSI) or requiring proprietary and expensive hardware for the SAN fabric (like Fiber Channel) it takes advantage a all many features of Ethernet such as multi-path awareness and the ability to "selfishly flood" an available link to increase performance. In fact, these features are why AoE can easily outperform iSCSI and are more comparable to FC speeds.
Raw Ethernet is a very effective transport medium, especially when using jumbo frames, and multiple paths in a switched environment (add another GigE link between initiator and target and AoE automatically saturates that link and increases performance). This could hardly be considered a fractional performance gain, IMHO...
USB 3.0 will be great for DAS... don't immediately see it replacing SAN storage protocols anytime soon.
I've been to Coraid and use AoE
USB is 3.2 Gbit/s
Coraid runs AoE over 10Gbit.
Decoupling the disks from the server has its uses. You can choose whichever disks you like as your RAID targets, from any host on the LAN.
If you can have 2 x 10Gbit interfaces in your machine you can get 6 disks to saturate it, even from 20 x 1 Gbit devices if you like.
You can build more capacity than any endpoint can handle.
If you are routing your SAN you have mis-architected your environment. You can't complain about a solution saturating your switches and advocate routing your block layer in the same argument.
Terrible - F.
What a waste of time
$1000/iSCSI port who are they kidding? The more 10 Gbe becomes adopted in the data center, the less this protocol will be relevant.
- Just TWO climate committee MPs contradict IPCC: The two with SCIENCE degrees
- 14 antivirus apps found to have security problems
- Feature Scotland's BIG question: Will independence cost me my broadband?
- Apple winks at parents: C'mon, get your kid a tweaked Macbook Pro
- FTC to mobile carriers: If you could stop text scammers being jerks that'd be just great