Gridstore bigs up hyper-converged box: Our storage nodes 'know' what Hyper-V needs Gridstore has introduced hyper-converged appliances for Hyper-V, including an all-flash variant. The startup has so far produced scale-out Storage Nodes that are aware of Hyper-V VM needs. Its Hyperconverged Appliance (HCA) complements these, we're told, and take Gridstore into competition with Maxta, Nutanix, Scale Computing …
@Tom Maddox
I don't doubt you've not met people who run Hyper-V for production but that may just be the circles you move in and the people you know as there is quite a lot of it out there.
In larger enterprises, VMware still rules the roost but some of that is no doubt due to a large existing investment. (Not saying that new deployments wouldn't also choose VMware!)
This isn't a VMware vs Hyper-V argument, of course, but there are most definitely plenty of product Hyper-V deployments out there
MS is even jumping on the convergence bandwagon with the next Server platform supporting something like this (can't comment yet on exactly how, I'm still downloading the ISO). But it looks like MS wants to eat some of Nutanix and their competitions' lunches.
See this for info http://technet.microsoft.com/en-us/library/dn765475.aspx (Storage Replica). I'm guessing it'll be file-based and synchronous commit, but until I build one...
I am very suspicious of the claims from this company. First of all erasure coding type of method to spread the data to multiple nodes for writes is going to result in large latency. The primary advantage of flash is very low latency and if you kill that advantage, why use flash ? Secondly, it is not clear if the product allows same data to be accessed and modified from multiple computers or if it only allows data to be spread to multiple computers but accessed & modified from only one location. If data block is spread to multiple machines then modifying that data requires it to be read from all those machines and then re-write the modification back to all those machines. If you allow modification of such data from multiple different computers then complexities increase many folds.This will cause lot of overhead and extra latency. How can they beat all flash array performance with this kind of complexity ?
I am not sure if this type of storage can ever be used as primary storage. This may be useful for backup but who would want to use all flash storage for backup ? It doesn't make sense. This company is trying to prop-up a very basic and commonly available technology of erasure coding by giving it a different name. Unfortunately this type of misleading advertisement that doesn't disclose full details has become quite common in industry. This is very unfortunate and I am not sure why a good website like register is also falling for such cheap tactics.
I think you didn't understand my comment at all. Sending all reads and writes in parallel to all nodes doesn't necessarily reduce your latency. In fact it is likely to increase it because now you have to wait for successful completion of write/read at "n" nodes. Think of a simple case of modifying a byte of data in a block which is very common operation in primary storage. You need to first read that block from "n" machines. Apply erasure coding to reconstruct the data and then modify the data. After that you have to do erasure coding again and then store the data back to "n" systems. Imagine doing this hundred thousand times per second. Your CPUs are going to get overwhelmed and latency is going to jump tremendously.
You can't compare this to pNFS. It does not do erasure coding.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
