Ramnode doing NVMe now

greattomeetyou

Nick_A said: We've seen up to 2.1GB/s dd within test guests on an empty node. fio writes just under 3000MB. Obviously, speed will vary depending on node usage, caching, etc. I would just expect it to be generally faster than standard SSD.

Is server NVMe the same as desktop NVMe in single drive condition? or does server NVMe have extra PCI lanes to make it faster?
Can server have raid on NVMe to double R/W performance?

Harambe

greattomeetyou said: Can server have raid on NVMe to double R/W performance?

I'd assume they're running them in RAID 1 or 10

jackb

@greattomeetyou said:

Nick_A said: We've seen up to 2.1GB/s dd within test guests on an empty node. fio writes just under 3000MB. Obviously, speed will vary depending on node usage, caching, etc. I would just expect it to be generally faster than standard SSD.

Is server NVMe the same as desktop NVMe in single drive condition? or does server NVMe have extra PCI lanes to make it faster?
Can server have raid on NVMe to double R/W performance?

I doubt it will double single drive performance (since it's so high - I suspect both hardware and software raid would have a hard time making use of 2x performance). But yes, you can use nvme drives in raid and I'd be surprised if ramnode weren't.

FHR

@Clouvider said:

@FHR said:

Nick_A said: caching

You are caching NVMe? Is it even worth it?

Ram’s still faster, to start with ;-).

Much more expensive though, especially now.

greattomeetyou said: does server NVMe have extra PCI lanes to make it faster

Server grade NVMe can be actually slower. But they have much greater durability (DWPD)

LosPollosHermanos

So are people not bothering with SW RAID on this stuff? What happens when they fail? Do they just turn into read only allowing a backup? Of course offline backups are done just like with HD's.

Nick_A

NL is available now - I believe some of you were waiting for that.

Adam1

Would one of these generally help with DB performance? currently on a ramnode openvz premium plan and I/O is sometimes a bottleneck. will rmanodes nvme iops be better than whatever ssd's are in their (presumably) sata ssd vps's?

Nick_A

@Adam1 said:
Would one of these generally help with DB performance? currently on a ramnode openvz premium plan and I/O is sometimes a bottleneck. will rmanodes nvme iops be better than whatever ssd's are in their (presumably) sata ssd vps's?

Yep!

Jones

@Nick_A

LA is available now？？？

Nick_A

@Jones said:
@Nick_A

LA is available now？？？

Not yet, sorry.

MikeA

@Adam1 said:
Would one of these generally help with DB performance? currently on a ramnode openvz premium plan and I/O is sometimes a bottleneck. will rmanodes nvme iops be better than whatever ssd's are in their (presumably) sata ssd vps's?

It does, but it probably wouldn't be noticed with most apps.

willie

I had Ramnode OpenVZ 128MB premium SSD plan for many years (finally cancelled because too many idlers) and it was always very speedy when I had it. Definitely fast enough for any sort of light access database usage. For heavier duty yeah you want something heavier duty.

sin

I picked up one of their NVMe VPSes last night and the performance is pretty damn good! Ramnode just enabled cpu passthrough (so I could have aesni) for me and the cpu cores are 3.7GHz.

Here's a quick FreeBSD diskinfo test:

root@freebsd:~ # diskinfo -cti /dev/vtbd0
/dev/vtbd0
    512             # sectorsize
    26843545600     # mediasize in bytes (25G)
    52428800        # mediasize in sectors
    0               # stripesize
    0               # stripeoffset
    52012           # Cylinders according to firmware.
    16              # Heads according to firmware.
    63              # Sectors according to firmware.
                    # Disk ident.

I/O command overhead:
    time to read 10MB block      0.013958 sec   =    0.001 msec/sector
    time to read 20480 sectors   2.374770 sec   =    0.116 msec/sector
    calculated command overhead         =    0.115 msec/sector

Seek times:
    Full stroke:      250 iter in   0.022587 sec =    0.090 msec
    Half stroke:      250 iter in   0.029474 sec =    0.118 msec
    Quarter stroke:   500 iter in   0.062339 sec =    0.125 msec
    Short forward:    400 iter in   0.047338 sec =    0.118 msec
    Short backward:   400 iter in   0.050634 sec =    0.127 msec
    Seq outer:   2048 iter in   0.111607 sec =    0.054 msec
    Seq inner:   2048 iter in   0.106352 sec =    0.052 msec

Transfer rates:
    outside:       102400 kbytes in   0.094869 sec =  1079383 kbytes/sec
    middle:        102400 kbytes in   0.093738 sec =  1092406 kbytes/sec
    inside:        102400 kbytes in   0.089908 sec =  1138942 kbytes/sec

Asynchronous random reads:
    sectorsize:    137690 ops in    3.000539 sec =    45888 IOPS
    4 kbytes:      134037 ops in    3.000558 sec =    44671 IOPS
    32 kbytes:     127723 ops in    3.000570 sec =    42566 IOPS
    128 kbytes:     90931 ops in    3.006833 sec =    30241 IOPS

yomero

sin said: Here's a quick FreeBSD diskinfo test:

Pretty interesting tool! That's new here Thanks!

sin

Here's some tests done from Debian 9.4 (using backported kernel):

root@ram:~# ioping . -c 10
4 KiB <<< . (ext4 /dev/vda1): request=1 time=110.9 us (warmup)
4 KiB <<< . (ext4 /dev/vda1): request=2 time=231.7 us
4 KiB <<< . (ext4 /dev/vda1): request=3 time=219.0 us
4 KiB <<< . (ext4 /dev/vda1): request=4 time=246.0 us
4 KiB <<< . (ext4 /dev/vda1): request=5 time=211.4 us
4 KiB <<< . (ext4 /dev/vda1): request=6 time=243.3 us
4 KiB <<< . (ext4 /dev/vda1): request=7 time=232.8 us
4 KiB <<< . (ext4 /dev/vda1): request=8 time=213.4 us
4 KiB <<< . (ext4 /dev/vda1): request=9 time=228.9 us
4 KiB <<< . (ext4 /dev/vda1): request=10 time=216.7 us

--- . (ext4 /dev/vda1) ioping statistics ---
9 requests completed in 2.04 ms, 36 KiB read, 4.40 k iops, 17.2 MiB/s
generated 10 requests in 9.00 s, 40 KiB, 1 iops, 4.44 KiB/s
min/avg/max/mdev = 211.4 us / 227.0 us / 246.0 us / 11.9 us

root@ram:~# ioping -RL /dev/vda1

--- /dev/vda1 (block device 23.0 GiB) ioping statistics ---
14.6 k requests completed in 2.96 s, 3.56 GiB read, 4.93 k iops, 1.20 GiB/s
generated 14.6 k requests in 3.00 s, 3.56 GiB, 4.86 k iops, 1.19 GiB/s
min/avg/max/mdev = 166.0 us / 202.9 us / 4.44 ms / 106.8 us
root@ram:~# ioping -R /dev/vda1

--- /dev/vda1 (block device 23.0 GiB) ioping statistics ---
25.9 k requests completed in 2.96 s, 101.1 MiB read, 8.76 k iops, 34.2 MiB/s
generated 25.9 k requests in 3.00 s, 101.1 MiB, 8.63 k iops, 33.7 MiB/s
min/avg/max/mdev = 34.8 us / 114.2 us / 2.65 ms / 34.5 us

root@ram:~# hdparm -Tt /dev/vda1

/dev/vda1:
 Timing cached reads:   22950 MB in  1.99 seconds = 11512.35 MB/sec
 Timing buffered disk reads: 5334 MB in  3.00 seconds = 1777.46 MB/sec

root@ram:~# wget https://freevps.us/downloads/bench.sh -O - -o /dev/null|bash
Benchmark started on Sun May 20 14:42:33 EDT 2018
Full benchmark log: /root/bench.log

System Info
-----------
Processor   : Intel(R) Xeon(R) CPU E3-1240 v6 @ 3.70GHz
CPU Cores   : 2
Frequency   : 3696.048 MHz
Memory      : 1997 MB
Swap        :  MB
Uptime      : 7 min,

OS      : Debian GNU/Linux 9
Arch        : x86_64 (64 Bit)
Kernel      : 4.16.0-0.bpo.1-amd64
Hostname    : xxx


Speedtest (IPv4 only)
---------------------
Your public IPv4 is x.x.x.x

Location        Provider    Speed
CDN         Cachefly    106MB/s

Atlanta, GA, US     Coloat      111MB/s 
Dallas, TX, US      Softlayer   52.1MB/s 
Seattle, WA, US     Softlayer   25.0MB/s 
San Jose, CA, US    Softlayer   28.9MB/s 
Washington, DC, US  Softlayer   45.3MB/s 

Tokyo, Japan        Linode      4.80MB/s 
Singapore       Softlayer   8.12MB/s 

Rotterdam, Netherlands  id3.net     7.05MB/s
Haarlem, Netherlands    Leaseweb    34.3MB/s 


Disk Speed
----------
I/O (1st run)   : 1.8 GB/s
I/O (2nd run)   : 1.8 GB/s
I/O (3rd run)   : 1.8 GB/s
Average I/O : 1.8 MB/s

and last but not least, here's a Geekbench

I really like this Ramnode NVME VPS, def a keeper :-).

Sofia_K

OMG cpu cores are 3.7GHz! that's terrific. Will try them.

sin

@Sofia_K said:
OMG cpu cores are 3.7GHz! that's terrific. Will try them.

Yeah it's not just the disks that are blazing fast, they give you some fast cores too!

sin

Whelp I ended up migrating my websites on a VULTR NJ VPS to my Ramnode NVMe VPS in ATL and I couldn't be happier! The performance is really freaking good for $12/Month and the only issue I came across was for some reason my Ramnode IPv6 can't access any OVH IPv6 addresses (discovered this when I tried using ipv6-test.com which is hosted at OVH and then tried doing mtr -6 and ping6 to and from my Ramnode ATL and Kimsufi server) but it works fine for every other IPv6 network.

I'm probably going to pickup a Ramnode NVMe VPS in NY next and maybe setup a DragonFly BSD server there.

Nick_A

@sin send us a ticket with the IP range you're trying to reach. I bet I know what the problem is.

sin

@Nick_A said:
@sin send us a ticket with the IP range you're trying to reach. I bet I know what the problem is.

Done, Ticket #190102 - thank you :-).

JoeMerit

@sin what plan did you come from at Vultr ?

sin

@JoeMerit said:
@sin what plan did you come from at Vultr ?

I was using a few $10 vps plans at VULTR and was in the middle of setting up a new one there so I could condense some of them when I saw this post here that said Ramnode had new NVMe plans. I hadn't of used Ramnode in a long time so I figured I would give their new nvme vpses a try and I was super impressed with the performance so decided I would setup my new webserver on a Ramnode NVMe VPS instead.

I usually use FreeBSD on most of my servers (currently testing out OpenBSD and DragonFly BSD on some others) so I was happy to see Ramnode had all the latest BSD isos available too :-).