The Xen vulnerability announced October 1 caused AWS, Rackspace, and SoftLayer to patch and reboot many cloud instances in the preceding week. We created a benchmark to measure the price vs performance of public cloud providers using Apache Cassandra.
We ran the test multiple times in each provider’s environment to eliminate temporary variability.
As far as raw performance goes, the workload performed best on the AWS c3.2xlarge and the RSC 30-Performance instance types. Latency measured above is the 95th percentile for the amount of time an insert operation took from the client. Bringing the price per hour of each instance type into the equation paints a clearer picture of what you’re actually getting. We’ve run the benchmark against each of the providers 4-core instance types to compare the price vs. The top performers have remained the same, the AWS c3.2xlarge and Rackspace 30-Performance are still on top. Interesting stuff, I’ve been looking for some Cassandra performance tests on the new AWS C3 instances.
Do you recall if your C3 instances were using SR-IOV or as AWS calls it “enhanced networking”? I would also like to see this comparison updated with the last reduced prices in both GCE and AWS. Rackspace Hosting just announced its own high-performance, flash-based server instances on its public cloud, and Amazon Web Services has answered with EC2 compute cloud instances that have the latest Intel Xeon processors, big gobs of memory, and flash storage to support the most demanding applications. The new C3 instance types on the EC2 compute cloud are based on two-socket servers using Intel's Xeon E5-2680 v2 processors. Werner Vogels, CTO at Amazon, introduced the new C3 instances during his keynote at the re:Invent conference that AWS is hosting in Las Vegas this week, saying that it was the "absolute highest performance instance" that AWS has.
The Performance Cloud Server instances that Rackspace announced last week have larger amounts of memory and flash storage for a given number of virtual CPUs than the C3 instances, but they are based on the earlier "Sandy Bridge-EP" Xeon E5-2600 v1 processors. Don’t forget that the relative performance of those high-core Xeons is much lower than the equivalent i7. We wanted to go beyond the anecdotes, tweets, and blogs to survey cloud users and find out how they actually fared during the reboot: Did they experience downtime of their applications? By utilizing Cassandra, we are able to demonstrate how a real-world distributed application performs in each cloud provider’s environment. Using the cassandra-stress tool from a separate load generating node, we generated a write-heavy workload.
This means that each write was synchronously written to a single node, and then asynchronously replicated to the other two nodes in the cluster. On average these instance types performed nearly 40% faster than GCE, and almost twice as fast as the c1.xlarge.


This measurement is the total round trip time for an insert to complete with a consistency level of ONE. Below we compare the cost per hour of each instance type normalized on one million operations.
In order to get an even more comprehensive picture of how they match up, we would want to create tests for other key metrics. Using bigger (or smaller) instances could yield different price vs performance when scaling out a distributed system. The new 4-core instances AWS c3.xlarge and Rackspace 15-Performance perform on par with the GCE 8-core n1-standard-8.
The AWS instances can improve their scores for a little effort on your part and no additional cost.
New prices, and particularly the new Cassandra 2.0 deployment tuned for Google Cloud make me wonder how much weight there is to the recent benchmark published by Google. Please use newer version of your browser or visit Internet Explorer 6 countdown page for more information. Vogels added that the inter-instance latencies have been "brought down much lower" and it allows customers to run the highest performance HPC applications.
That two-socket Xeon E5-2680 v2 server has 20 cores and 40 threads, so it looks like a lot of the processing capacity – but not all of it – is being burned up in the c3.8xlarge instance type.
A virtual cluster with 8,192 cores was also tested and delivered 163.9 teraflops sustained on the Linpack test.
Rackspace has not said which Xeon E5 it is using, so it is not possible to gauge relative performance.  Rackspace is charging less than AWS for skinny slices of its high-performance cloudy servers, but considerably more for larger slices of the same machines. The workload we ran against each Cassandra cluster was primarily focused on testing CPU and network. The stress tool measured the operations per second that it was able to perform against the cluster.
The AWS c3.2xlarge and RSC instance types are dead even, coming in at just under 2ms per operation. The AWS c3.xlarge beats out the group, showing that for this workload using the smaller instance type costs less per operation. Combining this benchmark with a read-heavy workload should give us a more complete picture of price vs performance overall.
And AWS is letting Turbo Boost run as well as giving customers access to the AVX vector math units on the chips, too.
HyperThreading is available on these processors and is turned on, and each virtual CPU is running on one of those virtual threads. What would be really interesting is to see a bakeoff between these two instance types on a variety of benchmarks. What were their lessons learned?Our survey was conducted on Oct 2-3, 2014, (after all the reboots had completed) with 449 respondents.


Rackspace’s 15GB-Performance instance does much better than the 30GB-Performance, coming in 3rd overall. As it has been doing for certain kinds of instances, AWS is also talking about what particular processor is underneath the C3 instances so companies can tune up their applications. We got responses from 349 AWS users, 66 Rackspace Public Cloud users, and 42 SoftLayer Virtual Server users. For GCE the difference is negligible, both the 4-core and 8-core instance types are about even when comparing cost per operation. In fact, it is hard to imagine a workload that would not benefit from such a configuration.
We also received 74 responses from organizations that used Xen in their internal data centers.
Note that some respondents used multiple clouds, so the total adds up to more than 449.AWS Users Had Less DowntimeThe most critical measure of the reboot was the impact on customer applications. AWS came out on top with 51 percent respondents who use AWS reporting no application downtime as a result of the reboot, and another 21 percent reporting less than 5 minutes of downtime. The percentage of instances slated to reboot varied for each individual AWS user depending on the instance types used and luck of the draw.
Thirty-nine percent of survey respondents using AWS saw less than 10 percent of their instances slated to reboot while 51 percent saw more. Because the AWS reboot did not affect all instance types or all instances, AWS users were able to relaunch instances ahead of the reboot (29 percent) and move resources to unaffected instance types (12 percent). In addition, a large number of AWS users (43 percent) were using multiple availability zones (AZs).
As a result, 48 percent of Rackspace users and 39 percent of SoftLayer users took no preventative action, as compared to 20 percent of AWS users.Most Cloud Users Weathered Reboots WellCloud users were given relatively short notice by cloud providers. AWS and Rackspace users started receiving maintenance notices roughly 36 hours in advance of the first reboot, although the Rackspace notices went out late Friday evening to warn of reboots starting Sunday morning. However, 41 percent of Xen users reported that the patching process created a lot of extra work, a higher percentage than all of the public cloud users.Cloud Users Remain UndauntedThe broad majority of cloud users are likely to continue using their provider despite the reboots.
Only 10 percent of AWS users, 20 percent of Rackspace users, and 29 percent of SoftLayer users are less likely to use their particular cloud provider.
Similarly only 16 percent of Xen users reported they are less likely to use Xen.Cloud Reboots Were a Wake-Up CallCloud users plan to continue using cloud despite these reboots, with only 5 percent expecting to use cloud less. But their recent experience is pushing many cloud users to implement additional cloud and IT best practices.
Cloud users intend to avoid futured downtime by making such changes as creating a plan (37 percent), implementing redundant architecture across AZs (35 percent) or regions (27 percent), employing more automation (30 percent), and leveraging multiple cloud providers (26 percent).



Free storage software download
Download panda cloud antivirus free edition 2014 review
Install owncloud fedora 20 000


Comments

  1. 09.04.2014 at 16:12:57


    Performance, and scalability ??especially based on confidentiality, integrity and secure cloud storage. Recommend to friends feature.

    Author: Elektron
  2. 09.04.2014 at 15:55:30


    Easy to find what you virtual desktop within the cloud the 250GB storage.

    Author: DeatH
  3. 09.04.2014 at 11:20:31


    You can simply restore your photos, and other iPhone files enough to sort out your.

    Author: Ayshe
  4. 09.04.2014 at 21:31:21


    You could get (usually more than the free plans.

    Author: isk