New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
@imgmoney
Instead of trying to solve your problems (do they really exist and have you analyzed them?) by throwing money at hardware and software it might be more promising to (a) properly analyze your problem, and (b) to throw some money at someone who has the knowledge and expertise to guide you towards a sensible solution.
(No, this is not an attempt to apply)
An update, site is up and running uncached version on "AMD Ryzen 7 3700X" and everything in 1 server ( nginx + php-fpm + mariadb + redis + elasticsearch )
It is 5 time faster without caching.
I wonder what made it super fast?
1) CPU have little higher benchmark
2) IOPS is 25 times higher than previous setup since it is NVME RAID 1 with 4xSSD RAID10
Any inputs?
1) Should I downgrade the CPU to "AMD Ryzen 5 3600 " and use 4x NVMe in RAID10?
2) Should I upgrade to better CPU and use NVMe and use the same RAID1and form cluster?
I am going to benchmark it on "AMD Ryzen 9 3900" and "AMD EPYC™ 7502P" as well.
Is there any difference in NVMe SSD Datacenter Edition and NVMe SSD, in benchmark I am seeing less IOPS for datacenter edition.
And right now write and read speed is 1GBps+ , if I use RAID 10, I can get 4GBps. Will it increase the app performance? If I want to utilize the higher IOPS, do I also need to have high capacity port of 1gbps+
Thanks
That would imply you were limited by IOPS, although only objective testing will verify that. You should run identical benchmarks on the new 3700X machine and the old E5 one, and see which component is the bottleneck, and then change your machine depending on the results.
Datacenter edition will be able to sustain that benchmark, while the normal NVMe will get throttled after a while. Also, more endurance on the DC SSD (not that it matters much in this case).
If you were limited by disk IOPS, yes, that would increase performance. Highly doubtful that you would need a higher port speed, although you should check it, just to be safe.
I think you should see some metrics on your current setup first. Try with
glancescommand on all nodes. Have a look what resources have high usage (CPU, iops, memory, or bandwidth), and that is what you need to upgrade. If you don't find anything, it indicates that maybe your set up needs to be adjusted to use more resources, something like increasing fpm workers or adjust mariadb memory cache.Also take a look on access log statistics. I have one site that have high mariadb cpu usage and upgraded the machine to 32 core cpu. It turns out that the site has many bot. After I blocked all bot IP, the site can be run on 6 core with relatively low cpu usage.
If you use proxysql, it also can become bottleneck to the database because it needs to differentiate read write traffic. Try to use mariadb galera cluster with multimaster setup and haproxy to do tcp balancing.
If you have site with high traffic, use only the NVMe datacenter edition. It has slightly lower iops, but stable performance. Believe me you don't want to investigate in the future just to find out that your disk iops is throttled and ruin your overall cluster performance.
@akhfa
Among these two which will have better response time
1) proxy SQL - one master + one slave , no gtid validation
2) tcp load balancing - Galera cluster multi master setup
I read multi master setup aka multiple writers is slow when comparing with simple master slave replication.
I believe Galera replication use multiple checks to make data consistency and many other logics. Won’t it make the response time slower than simple master slave setup ?
The best answer is when you try and bench yourself. I don't have Magento cluster, so I don't know what kind of database load it is. In my case, proxysql becomes the bottleneck. Use something like apache bench or https://github.com/tsenart/vegeta
But again, if your bottleneck is not on your proxysql, I think it is not worth to be tried yet, and focus on your real bottleneck first
Shit, this thread escalated quickly.
@imgmoney thanks for sharing benchmarks. Please post benchmarks of 7502 and 3900x when you get them.
Hello! We can suggest you
Amsterdam
DELL R420/2xE5-2430/32GB(4x8GB PC3-10600 RDIMM)/2x3TB SATA/2x240GB SSD EDGE
+100mbps= 120$/mo
https://inxy.com/dedicated-servers/
@dodheimsgard
Here are the benchmark for the same
https://browser.geekbench.com/v5/cpu/2985833
https://browser.geekbench.com/v5/cpu/2990641
Well, but our application performs well in this spec
AMD Ryzen 7 3700X
64GB ECC RAM
2x1TB NVME
Why the app didnt perform well in other two specs? I am confused as the other two CPU are way too higher.
AMD Ryzen 9 3900 - Performance degrade by 5%
AMD EPYC 7502P - Performance degrade by 100%
Why?
I have ordered these machines too
AMD Ryzen 5 3600 Hexa-Core
4x512GB NVMe - RAID 10
64GB ECC Ram
and
AMD Ryzen 5 3600 Hexa-Core
2x512GB NVMe - RAID 1
64GB Non ECC Ram
Other questions
1) Do ECC and non ECC Ram plays major role if I use it for HAProxy? Or for Web Servers? I save 5euros, every penny matters.
2) Since some one told Datacenter NVMe are better as normal NVMe may throttle later. So I have gone with RAID10, am I able to overcome that issue or I am fine with normal NVMe as those config doesnt have Datacenter NVMe
Must likely because your application need higher clock speed so 7 3700X base clock speed is higher than the others ( I'm not talking about the turbo ! )