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
Either that or it's being run when other folks are running theirs. Seems to happen occasionally on the main site when folks are posting reviews.
@kiloserve the issue with the slow VPS is random IO, not speed. ioping showed results in the 150-500ms range on the slow VPS and 0.1-0.4ms on the fast VPS.
I just use vzfree and check how much oram is under (Swap:)
This should also affect sequential write speeds as well. There is disparity between random and sequential; however, they should be close and you won't get a poor sequential speed and then get blazing fast random writes or vice versa.
50 MB/s is an average write speed for DD. It's random write speeds should be average, random writes won't be poor and it won't be great either.
Conversely, if I have a DD score of 10MB/s, my random writes will also be poor.
Or if I have a DD score of 300 MB/s, my random write scores will also be excellent.
If anybody actually wants to test it out, I do have 2 servers I can provide you for a few days.
One is off our XenPV nodes and scores about 250-300MB/s in DD.
The other is another providers box which scores around 100MB/s in DD.
I can give you access to both boxes and you can run random write I/O tests and post the results here.
My guess is that under any random or sequential disk benchmark the 250MB/s box beats the 100MB/s box.
If you take me up on the offer, you must agree not to disclose the other provider's name/IP/etc, can't be a VPS provider and you must be a regular LEB/LET poster here in good standing.
Better random IO, but low dd score:
Decent dd score, but bad small file reads:
First one is a over 10 year old PC with basically nothing running, second is a VPS which got ~130mb/s when I first got it.
Thanks for running the tests Doc, interesting indeed.
This one is understandable as the disk I/O is dedicated rather than in a multi-user environment. If you simulate a multi-user environment by adding another disk process like copying a directory with a bunch of random files while running your tests, you will get a more accurate representation of DD on a VPS.
I am guessing with simultaneous disk access (like you have in a VPS) you will see both DD scores and read scores get more in line with your VPS testing.
On your second VPS, when you were getting 130MB/s, wasn't the random I/O great back then? I think the two go hand in hand, when the sequential DD score drops that much, so will the random scores.
Just for comparison, here's another VPS test. Similar DD scores with similar random reads.
The DD score was slightly higher and the random reads are also slightly higher.
And here's a comparison on a node with almost 20 VPS on it.
As you can see here, the DD score correlates with the random reads as well. As the DD score goes down, I suspect the random read times will increase.
I never ran ioping on the VPS when I first got it, but it was defiantly much faster when I first got the VPS.
This is a VPS, with nice and stable ioping result and ok dd result. Much better disk performance than the VPS with a 64MB/s dd result.
The disk reads are excellent but the write speed is slower. So if a user wants just excellent read speeds, this is a good choice. But at the same time the write speeds are slower. I can't really say that slower write speeds is better but more of a personal choice depending one whether reads or writes are more important.
The other VPS can write faster, this one can read faster; but both are average VPS and considerably slower than a VPS that posts 283MB/S in DD
Let's try some small writes:
First is the VPS with 64MB/s DD and the second is the VPS with the 59MB/s DD.
Here's the result from the 283MB/s DD server, as you can see it is far superior to the two average speed servers as the speed indicates.
The two average servers also seem to write at about the same max speed.
The trend is still the same, you get 2 average DD scores and you get average results. You take a high DD score server and it outperforms the average DD score servers.
The ~60 MB/s average DD server is not going to beat a 283 MB/s DD server.
Small writes are less accurate than big writes imho.
Maybe a good test can be concurrent small writes
Folks, please take a look at this:
Kernel: 2048.00M 20.43M 2027.57M
Allocate: 4096.00M 677.49M 3418.51M (2048M Guaranteed)
Commit: 2048.00M 366.80M 1681.20M (51.1% of Allocated)
Swap: 2.25M (0.6% of Committed)
Is it good or bad?
Sorry, If I bumped such an old thread, but I'm curious with my vzfree results, which are:
Total Used Free Kernel: 2048.00M 10.29M 2037.71M Allocate: 1024.00M 497.64M 526.36M (512M Guaranteed) Commit: 512.00M 259.51M 252.49M (50.1% of Allocated) Swap: 0.11M (0.0% of Committed)
From what I understand, my node has been swapping, but considering the number of memory that get swapped (0.11M) relative to the amount of RAM allocated (0.0-ish %).
Does it matter? I mean, will it cause any performance degradation?
I don't think so. Get worried when more RAM starts to move to the swap
Hey, thanks for commenting Yomero
Well, it goes back to 0.00 M for now, I guess I was just being paranoid. No?
Would love to hear from other LEB master too.
LOL, nah, I am not a master :P but thanks.
I remember when I had my hostrail boxes and that were moving more than the half of the RAM used to the swap :S
Lol, yeah. My first ever VPS was from HostFail.
I don't believe it is actually possible to detect if a provider is "really" overselling, Measuring your results on the Disk I/O, Net Speed, Etc.. Is not a accurate way.
Your provider will measure the nodes on the CPU load, This determines how many clients a single node can handle and what would be the CPU threshold on a node.
You can still make a profit on nodes without majorly overselling, Even if your a budget provider.