I admittedly know nothing about AWS, but is it really an "instance" if it consumes the entirety of the resources on the physical computer? If you're running a critical database, you probably aren't running temporal instances of that service, are you? You're just leasing hardware from Amazon at that point, right? How can you rollover failures or transfer instances for services that run entirely in memory?
AWS chucks 2TB X1 instances at SAP memory hogs
Amazon’s released AWS instances packing 2TB serving mega memory-hungry workloads such as SAP HANA. The cloud provider today uncorked its X1 instances, which were first announced in October. X1 instances use four Intel Xeon E7 2.3GHz processors with 10Gb per second of dedicated bandwidth and large L3 caches targeting high- …
COMMENTS
-
-
Thursday 19th May 2016 19:13 GMT sixoseven
An instance is an instance. There's really not much more to think about. If you think an instance might fail, then you have a backup instance. You can have it in parallel, you could have it on hot-standby, you could have a smart data management workflow that breaks up your computing task determined by what's on disk and what's in memory. SAP HANA looks like a shardable k-safe database, thus the problem is solved with the database architecture itself, not with how you host it. I have an Oracle white paper that excoriates HANA's implementation and it has some legitimate points, but really the more you get deep into that debate, the more the simplicity of Amazon's Redshift becomes desirable.
-
-
Wednesday 18th May 2016 17:15 GMT Anonymous Coward
An x1.32xlarge instance has 128 vCPUs, 1952GB RAM, and costs $16 per hour. Pretty awesome.
I would guess they still run the instance inside a hypervisor, even if there is only one per box, so that they can send signals to the instance (e.g. shut it down), handle the virtual networking etc.
-
-
Wednesday 18th May 2016 18:16 GMT DonL
Re: $140,000 per annum
Indeed, or you can just buy your own server for around $100.000 and run it for 5 years saving you at least $400.000. It would take less than 9 months to reach the break-even point.
( http://www8.hp.com/us/en/products/proliant-servers/product-detail.html?oid=6636692#!tab=models )
But hey, CAPEX is bad and OPEX is good right?
-
Wednesday 18th May 2016 21:33 GMT Jon 37
Re: $140,000 per annum
But Amazon headline prices are for flexible capacity. If you know you're going to use the instance constantly, you can get a discount.
At the simplest level, if you commit to a 1 year term then you can pay $6700/month instead of $9600/month.
If you know you're going to want the server for 3 years then a one-off upfront payment of $98000 will cover it. Or if you don't want to pay so much upfront then you can pay roughly half the upfront amount ($52000) and the rest at $1400/month over the 3 year term, which comes to a total of $104000. Either way, the cost is roughly $100k, which is roughly what you say it would cost to buy the server yourself.
(Source for AWS pricing: https://aws.amazon.com/ec2/pricing/ set to the default US East region).
And the price for the server yourself doesn't include datacenter space or power or networking gear or Internet access, although the AWS price doesn't include Internet data transfer fees.
If you have a hardware issue with the server, with Amazon you can just spin the virtual machine up on another one of their servers, at no extra cost; you can't do that with your in-house server unless you spend money on a spare. Also, if you need a spare server due to a busy period, or for testing a software update, then you can always spin up another server on Amazon for a few days (at the spot price).
There are advantages to both in-house and in-cloud servers; but the cost difference isn't as much as you're making out.
-
-
-
-
-
Thursday 19th May 2016 04:40 GMT Simon Brady
Re: Must stop glancing at headlines
Well for all we know, there could be 2,046,820,352 ZX81s providing that 1952GB, with 15,990,784 clustered together for each vCPU (because not even Amazon with their "everything fails, all the time" design philosophy would trust those dodgy RAM expansion cartridges).
-
Thursday 19th May 2016 07:52 GMT David 132
Re: Must stop glancing at headlines
Indeed, one RAMpack wobble and your SAP data is toast.
Do Amazon charge extra if the ZX81 instances are in FAST mode vs. SLOW, I wonder?
(Sorry, I didn't mean to take this discussion off-topic. I shall now return you to the discussion about SAP for which you came here!)
-
-
-
Thursday 19th May 2016 14:50 GMT DNinjaDave
vCPU:Core ratio
Each X1 instance is powered by four Intel® Xeon® E7 8880 v3 (Haswell) processors and offers 128 vCPUs.
Those E7 8880 v3's have 18 Cores per CPU or 36 Threads so their vCPU count of 128 means they must be using hyper-threading to achieve the vCPU count. Or roughly 2:1 vCPU to Core.
Do their other offerings that give you 2 vCPUs for instance mean you have 1 physical core with HT or 2 physical cores I wonder?