10
votes
Accepted
Why do sequential writes have better performance than random writes on SSDs?
A reasonably concise explanation by Seagate (WayBack copy here) on how garbage collection is responsible for the difference in SSD performance for random versus sequential writes:
... the need for ...
- 80.3k
7
votes
Estimate FLOPS in Linux?
For ballpark-estimates:
Raspberry Pi 2: 299.93 * 10^6 FLOPS (source)
Raspberry Pi 3: 462.07 * 10^6 FLOPS (source)
GTX Titan Black GPU: 5.1 * 10^12 FLOPS (source)
Sunway TaihuLight: 93 * 10^15 FLOPS ...
- 297
5
votes
Accepted
Understanding ZFS NVMe benchmarks with FIO
You are probably running fio on a recordsize=128k dataset. When reading/writing 4K blocks you are going to see up to 32x I/O amplification.
To get maximum IOPs you can create a new dataset with ...
- 50.7k
4
votes
Anyone seen a meaningful SAS vs SATA comparison/benchmark?
So much misinformation here!
I can't imagine a meaningful benchmark between SAS and SATA and I wouldn't bother looking for one. This is like benchmarking a six passenger minivan versus high-speed ...
- 41
4
votes
Does Apache Benchmark allow intervals between each request?
You can use hey instead.
It has similar syntax as apache bench.
hey -n 50 -q 1 -c 1 http://www.google.com
sends 50 requests with a pause of 1 sec between.
- 151
3
votes
Apache ab - testing with 1000 concurrency
What you're actually seeing is an application that is "serializing" request processing- meaning, only one request is being processed at a time, regardless of the number of requests that are being ...
- 1,252
3
votes
Sysbench MySQL cannot connect
Create sbtest database and sbtest user.
If you're not going to supply command parameters to the sysbench command then you'll need to create the database and user that it expects when talking with the ...
- 165
3
votes
How can I verify my vserver provider offers SSD as promised?
There is no reliable way to determine whether a virtual machine's backing store is based on SSD or not.
The hypervisor presents a completely virtual disk device to the virtual machine, and no ...
- 248k
3
votes
Accepted
Litespeed vs Nginx high-traffic server performance?
What do you think about those benchmarks? 4x, 10x faster?
Well, the benchmarks are quite real - the benchmark is done using LiteSpeed webserver 5.4 (which is rather "new") which got a big overhaul ...
- 169
3
votes
Accepted
Root-causing vastly different performance on iozone O_SYNC benchmark for two HDD manufacturers
Regarding the measured 33x difference between your results, following up on our discussion in the comments, it turned out, that MegaCli64 -LDGetProp -DskCache -Lall -aAll showed that setup B had the ...
- 1,155
3
votes
Understanding this error: apr_socket_recv: Connection reset by peer (104)
Besides the answers here, I have read a lot of other ones:
Replace localhost by 127.0.0.1
Update the apache version (I have ApacheBench, Version 2.3 <$Revision: 1807734 $>)
Add -r (Then I get ...
- 297
3
votes
IOMeter - What values should I test with?
From a SQL Server perspective
On a SQL Server box you would preferably test the disks with the following parameters, depending on where you will be storing the MDF, NDF, LDF and TEMPDB files:
All ...
- 2,055
3
votes
Accepted
why read is faster when using O_DIRECT flag?
O_DIRECT is faster than a generic read because it bypasses the operating system's buffers. You are reading directly from the drive. There are a couple of reasons this could be faster though keep in ...
- 1,108
2
votes
Benchmarking hard disks on Mac OS X SL
Using file copy in general and dd in particular isn't the great way to measure I/O performance. This is a very good wrap up about WHY it's so.
https://barreto.home.blog/2014/08/18/using-file-copy-to-...
- 14.1k
2
votes
Accepted
How to configure Fio to a more accurate result of IOPs
Reading the man page yields
runtime=int
Terminate processing after the specified number of seconds.
time_based
If given, run for the specified runtime duration even if ...
- 99.6k
2
votes
How to find what's causing Apache/httpd to run at such high memory usage
You can try using lsof to read the files open by the apache process:
lsof -p PID
Checking the apache logs for errors that correspond to the timestamps of the spider crawl in your access logs is also ...
- 84
2
votes
How to disable perf subsystem in Linux kernel?
Disable the HAVE_PERF_EVENTS kernel option and recompile the Linux kernel.
Also, if you mentioned that it got flipped back to on, the. odds are good that there are more than one other kernel settings ...
- 989
2
votes
What are the best possible ways to benchmark RAM (no-ECC) under linux / arm?
Write a file into an existing tmpfs like /tmp with dd as wazoox suggested, but limit its size to less then half of your free memory.
First, find out how much memory is available:
> free -h ...
2
votes
Wordpress apache benchamrk CPU usage
Server load is not an absolute measurement. It depends on the Cores that are available on the system.
For example, in a 4 Core system, Load "4" is the same as Load "1" in a 1 Core system.
Since you ...
- 21
2
votes
Accepted
Strange IOPS performance on AWS R3.large & R4.large instances
Your constraint appears to be coming from the network limits on the instance type, not EBS itself.
There's some reading between the lines required, but the EBS Optimized Instances documentation ...
- 23k
2
votes
Why do sequential writes have better performance than random writes on SSDs?
Another explanation is that sequential I/Os are easier to coalesce at all levels. Generally you have less overhead when you send the same data but using fewer but bigger I/Os thus you can reach a ...
- 1,255
2
votes
Accepted
Where is my bottleneck: RAM, CPU or disk?
RAM usage is at ~73% with almost no swap space being used, and all your cores appear to be maxed out. So RAM size and speed are fine, and the bottleneck is certainly the CPU.
Based off your ...
- 824
2
votes
Will an E5-2660 be a major improvement over a Q9550 for a webserver?
The E5-2660 will be a improvement in CPU performance over the Q9550. It's a 4 year old newer chip and has 8 cores vs 4, and since your new system is a dual CPU you will have 16 cores, 4 times as many ...
- 3,033
2
votes
Accepted
how to benchmark dbms (sql and nosql) on s390x architecture (IBM mainframes basically)
Congrats on access to a Z system.
For the comparison of various databases I can only provide some general guidance. Here are some items to consider as you formulate your plan.
Atomicity - divide ...
- 190
2
votes
How to calculate log per second?
In logql you simply query your logs with
count_over_time({label="labelValue"} [1s])
That will give you the count of logs selected per 1s interval.
--
I love seeing our project being used in ...
- 21
1
vote
Accepted
Cluster using virtual machine software with 2 workstations
Depending what you want to achieve. Possible yes, but you likely have pathetic IO (which is generally a bottleneck for databases) and better make sure you split the VM's "strategically" because your ...
- 51.9k
1
vote
Accepted
How to calculate /get the UBER value from S.M.A.R.T measurements in SSDs
It turns out that the UBER (unrecoverable bit rate error) can be calculated based on the raw value of the attribute #ID 187, reported_Uncorrect.
Several manufacturers tend to use different names for ...
- 283
1
vote
How to circumvent siege concurrency socket select timed out
Please update the siege.
wget http://download.joedog.org/siege/siege-4.0.4.tar.gz
tar -zxf siege-4.0.4.tar.gz
cd siege-4.0.4
./configure
make
sudo make install
siege.config
You may need to issue ...
- 119
1
vote
Sysbench on Redhat 7 Errors with Mysql Testing - PANIC: unprotected error in call to Lua API (cannot open oltp: No such file or directory)
The solution after googling this was that it was looking for a oltp file, but it didn't find it. Apparently they moved it and it was never documented judging from what some of the github comments say. ...
- 239
1
vote
Better stream performance when transparent hugepages are on
Transparent hugepages are the thing that is causing this performance increase. The kernel in Cent 6 and Cent 7 are different by quite a lot. THP can optimize large contiguous memory patterns, and ...
- 7,136
Only top scored, non community-wiki answers of a minimum length are eligible