Tag Archives: lebre-x

LEBRE-X Bandit Host AMD EPYC VPS

Our second LEBRE-X review of the year goes to Bandit Host, a provider who specialises in offering AMD VPSes. They offer both AMD EPYC and Ryzen VPSes, but under review today is the Bandit Host AMD EPYC UT-S-NVMe-VPS. Let’s get into the specs:

  • Processor: AMD EPYC Processor
  • 2 vCPUs @ 1996.250 MHz
  • 2GB RAM
  • 25GB NVMe Disk
  • 1Gpbs Port
  • AES-NI: Enabled
  • VM-x/AMD-V: Disabled
  • Location: Ogden, Utah

With AES-NI enabled, there is no performance penalty on LUKS encryption using AES on the VPS. However, VM-x/AMD-V is disabled, which is not good news for those who wish to run nested virtualization solutions such as Proxmox. Nonetheless, this is not a deal-breaker for most people; nested virtualization offers are relatively uncommon.

We now turn to the results of our extended benchmarking, which we ran daily for 35 days from 13 December 2019 to 16 January 2020. All charts are interactive (i.e. you can mouse over the lines to examine each data point).

Geekbench 5 Test

Geekbench 5 offers a quick overview of CPU and RAM performance in a single score. The following chart shows the single and multi-core scores of the Bandit Host AMD EPYC UT-S-NVMe-VPS across our test period.

Bandit Host adopts a fair-share policy, but it can be seen that the VPS is both very high performing and the performance is very consistent for both single (mean score 614, SD 7.52) and multi-core (mean score 1177, SD 16.8) over the period of testing. The AMD EPYC is not as much as a performance beast relative to the Ryzen family, but it packs plenty of punch for most web applications.

What is very noteworthy is the very, very low standard deviation, which indicates an extreme level of consistency of CPU performance. This suggests that Bandit Host manages the host node really well. Also, the high number of cores in the EPYC could have contributed to the stability as well; with many more cores, processing can be distributed far more evenly resulting in much more stable performance across VPSes at any random point of time.

fio Random Read-Write

We use fio to benchmark disk performance because random read-write is more indicative of real-world use compared to sequential read-write, which normally generates impressive metrics but is a poor indicator of real-world performance. The following charts show the random read and write IOPS performance at 4k, 64k and 256k block sizes.

The NVMe disk clearly show its superior performance in terms of IOPS relative to other storage media in terms of random reading (mean IOPS 43128, SD 1579) and writing (mean IOPS 43138, SD 1565) at 4k block sizes. The standard deviation is very low, suggesting very consistent performance.

Speed is still excellent for random reading (mean IOPS 23372, SD 1008) and writing (mean IOPS 23424, SD 1008) at 64k block sizes. Once again, the standard deviation is really low, a very positive indication of consistent performance.

Finally, in terms of our toughest 256k block size random read-write test, the speed in terms of reading (mean IOPS 6868, SD 116) and writing (mean IOPS 6921, SD 118) was still very, very decent, with an ultra-duper low standard deviation.

The common thread across the fio test data is very, very stable and consistent performance. The AMD EPYC is not a brand new node, and to maintain this level of stability and consistency is a very positive sign of Bandit Host’s management of its host node. There was a weird outlier data point on 15th December, but that appears to be totally random considering all the other data points, so we can safely ignore that outlier.

If you need to host database intensive applications, the NVMe performance of Bandit Host’s VPS is more than sufficient to meet the requirements of most medium websites.

ioping Latency

We use ioping to measure how responsive the disk is to requests. The chart below shows the results of average response times to 30 requests.

Again, the disk response times (in microseconds) are excellent (mean 287.8, SD 42.2). Once again, with the exception of that outlier data point on 15 December, the overall latency is fairly stable and consistent.

iperf3 Speeds

We use iperf3 to measure outbound internet speeds (i.e. how fast the VPS sends data out to users) to our own collection of iperf3 servers. Keep in mind that there are many factors involved in transmitting data across networks, especially across vast geographical distances. Spikes are not unusual. Note that we present our data in megabytes per second.

We first present the chart for European locations.

Again, our internal Romanian test server’s (mean 63.0, SD 14.6) network isn’t very stable, but there clearly very consistently strong throughput to our Polish (mean 100.3, SD 18.3) and German (mean 110.0, SD 3.7) servers, except for a couple of outlier readings for the Polish server nearing the end of our testing period, which isn’t a major cause for concern.

We now present the chart for our North America servers.

We only have one American test server in Chicago for this round of testing, and the results are again one of very sustained and consistently high performing throughput to Chicago (mean 109.5, SD 4.6), except for that outlier blip on 15th December.

Finally, we present the charts for our APAC servers.

Bandit Host appears to have one of the best networks to Asia. Network performance to our Singaporean test server is particularly strong (mean 91.7, SD 15.3). The performance to Australia tends to be kind of spiky (which is actually not unusual when packets travel over vast distances) but the performance one of the better ones that we’ve seen to Perth (mean 22.1, SD 23.0) and Sydney (mean 33.4, SD 15.3).

Final words

If you’ve made it to this part of the review, you can already predict what we are going to conclude: consistently stable high performing VPS. Although the Bandit Host AMD EPYC is not the beast that Ryzen is, the EPYC is pretty decent and more importantly, you can expect it to perform when you need it (if you want to shatter benchmarks, Bandit Host has Ryzens too). If you are an Asian customer, the good news is that Bandit Host has one of the better networks across the Pacific to Asia, and coupled with reliable performance of their hardware, you can’t go wrong with Bandit Host, unless you are willing to be extorted by Asian providers (not their fault because bandwidth in Asia is generally expensive, but you get the drift).

tl;dr: You will sleep well every single night with a Bandit Host VPS

Note: we have reached out to Bandit Host for their response to our LEBRE-X review because we believe that our reviewed providers should also have the opportunity to respond to our observations. We will publish their response in full if we receive any.

Response from Bandit Host’s Management (23 January 2020):

First, thank you for taking your time and the effort required to perform these extended benchmarks! It is a fantastic service you are providing to the community!

Our mission and goal is to provide reliable and consistent performance across the board. Without those two items you are unable to fully utilize any service without worry. Ultimately, we want to provide a rock solid platform everyone can depend on for growth, whether that be a personal project or a business project. We appreciate the fact that this analysis reflects some of our core values.

Additionally, while AMD-V is showing Disabled on your VPS we have updated the flags on most current VPSes to include nested virtualization. All new VPS’s deployed by default have that flag enabled now.

At the end of the day, we’re here to understand what the community wants and see how we can best provide that in a way that everyone can grow and thrive. With that, we’d like to invite you to review and bench our East Coast, Washington DC based services once they are online in late February/Early March. Everyone else, stay tuned!

Thanks again!

Stephen Delacalzada-Delong
BanditHost

LEBRE-X: Nexus Bytes Ryzen NVMe VPS

The very first LEBRE-X review goes to our Nexus Bytes Ryzen NVMe reference box REBOX1. Readers will already be familiar with REBOX1 because it is the reference box for value comparison using the LEBRE methodology. However, although the LEBRE methodology provides a clear measure of value based on paper specifications, it cannot adequately reflect actual performance. Therefore, we’ve subjected REBOX1 to an extended period of benchmarking as outlined in the LEBRE-X methodology.

The detailed specifications of the VPS is as follows:

  • Processor: Ryzen 7 3700X 8-Core Processor
  • 2 vCPUs @ 3593.248 MHz
  • 2GB RAM
  • 30GB NVMe Disk
  • 1Gpbs Port
  • AES-NI: Enabled
  • VM-x/AMD-V: Enabled
  • Location: Germany

With AES-NI enabled, there is no performance penalty on LUKS encryption using AES on the VPS. Also, with VM-x/AMD-V enabled, nested virtualization is possible, allowing deploying of solutions such as Proxmox. Certainly a nice CPU feature set.

We now turn to the results of our extended benchmarking, which we ran daily for 33 days from 14 December 2019 to 15 January 2020. All charts are interactive (i.e. you can mouse over the lines to examine each data point).

Geekbench 5 Test

Geekbench 5 offers a quick overview of CPU and RAM performance in a single score. The following chart shows the single and multi-core scores of the Nexus Bytes Ryzen NVMe VPS across our test period.

Despite the fair-share policy of Nexus Bytes, it can be seen that the VPS is both very high performing and the performance is very consistent for both single (mean score 1143, SD 16.1) and multi-core (mean score 2122, SD 61.2) over the period of testing. This indicates that the host node is very likely not oversold and well-monitored, ensuring consistently high performance for your applications whenever you require.

fio Random Read-Write

We use fio to benchmark disk performance because random read-write is more indicative of real-world use compared to sequential read-write, which normally generates impressive metrics but is a poor indicator of real-world performance. The following charts show the random read and write IOPS performance at 4k, 64k and 256k block sizes.

The NVMe disk clearly show its superior performance in terms of IOPS. The Nexus Bytes VPS is extremely fast and stable in performance when it comes to random reading (mean IOPS 80811, SD 4951) and writing (mean IOPS 80899, SD 4956) at 4k block sizes.

Speed is still excellent for random reading (mean IOPS 30185, SD 7897) and writing (mean IOPS 30319, SD 7926) at 64k block sizes, although the variability of performance increases as evident from the higher standard deviation.

For our most gruelling 256k block size random read-write test, the speed in terms of reading (mean IOPS 9386, SD 2402) and writing (mean IOPS 9241, SD 2366) was still stellar, with some variability similar to the 64k block size test.

The overall conclusion from the fio results is crystal-clear: there is very likely no problem running applications that require strong IOPS performance. Assuming that the database is well-tuned, this Nexus Bytes VPS is likely to be able to deliver a consistently buttery-smooth experience, even for busy database driven CMS websites.

ioping Latency

We use ioping to measure how responsive the disk is to requests. The chart below shows the results of average response times to 30 requests.

The spikes in the chart may worry some of you, but there is actually no cause for alarm. The disk response times are measured in microseconds (mean 373.5, SD 126.6). A microsecond (μs) is one millionth of a second, so despite the spikes, you will not notice any real-world difference. Also, if the ioping results are too flat, it is indicative of very low disk activity (meaning the host node is very empty) or there is some caching mechanism going on.

iperf3 Speeds

We use iperf3 to measure outbound internet speeds (i.e. how fast the VPS sends data out to users) to our own collection of iperf3 servers. Keep in mind that there are many factors involved in transmitting data across networks, especially across vast geographical distances. Spikes are not unusual. Note that we present our data in megabytes per second.

We first present the chart for European locations.

We see generally consistent performance to our Polish server (mean 107.9, SD 6.7). The spikes to our Romanian server (mean 67.1, SD 14.3) is more of an issue with the network of our Romanian server. Generally, we see speedy network speeds in Europe.

We now present the chart for our North America servers.

You must be wondering about that huge dip in network speed to Utah (mean 95.0, SD 32.6) during the Christmas holiday period. We are not sure what happened during that period, but there were issues. We ran some MTR tests but they were inconclusive; packet losses were random along the entire path. We conclude that the results for that period is a freak anomaly. If we look at the speed to Chicago (mean 110, SD 2.87), the network speed is generally consistent. If we disregard that freak anomaly period, we can see that network speeds to North America can mostly saturate the promised 1Gbps port speed.

Finally, we present the charts for our APAC servers.

The see-sawing is not unexpected, given the vast distance between Germany and the APAC locations. The speeds to Singapore (mean 82.3, SD 29.0) is decent on average, although depending on the day and time, some bottle-necking may be experienced. If you are from Australia, tough luck. Both Perth (mean 15.1, SD 10.5) and Sydney (mean 17.3, SD 11.6) do not perform well.

Final words

It appears that the actual performance of our Nexus Bytes Ryzen NVMe REBOX1 lived up to expectation. There is ample evidence suggesting that Nexus Bytes has a very solid VPS management model. The CPU/RAM and disk performance are excellent AND consistent. It is no understatement to say that their VPS is likely to perform on par with or even better than low-end dedicated servers for most use cases, especially their beefier VPSes. We are impressed, and you should be too.

tl;dr: Shut up and take my money now!

Note: we have reached out to Nexus Bytes for their response to our LEBRE-X review because we believe that our reviewed providers should also have the opportunity to respond to our observations. We will publish their response in full if we receive any.

Response from Nexus Bytes’ management (Jan. 15th, 2020):

Thank you admin for sharing your detailed analysis. We are glad that this review validated our company’s philosophy: well-balanced nodes with healthy cpu : memory : storage ratio. Small hiccups may arise here and there; however, we try our level best to nip things in the bud and ensure our family members are receiving steady and consistent service, regardless of how much they paid.

We value our customers and treat them more as a family member and not just a mere customer. We are not flashy in our advertising nor do we offer unsustainable prices. If anything, we prefer to under-sell and over-deliver than the other way round, and we will continue to ensure that our family members who put their trust in us get nothing less than the best that we can offer.

Jay C., NexusBytes LLC