My review of BuyVM VPS and service

edited September 2020 in Reviews

Even though BuyVM is well known in the LE communities, new users join on a daily basis, and they might not be aware of them. People often take the time to write a review when they are pissed off, while positive experiences are underreported. I think it is time for a review, so here it comes.

Full disclosure: my review is spontaneous, unsolicited, and there is (AFAIK) no rewarding systems at BuyVM for writing a review. In case there are, @Francisco, I accept cash only ;-)

I have had a few Slices (BuyVM name for VPS with dedicated virtual cores, or CPU threads) over the last 2 years or so. These are the major operating periods I have had most recently:

  • 1x 8GB/$30 Slice in LV (Las Vegas) [Ryzen 3900X].
  • 1x 512MB/$2.50 Slice in LV, [Ryzen 3900X].
  • 3x 1GB/$3.00 Slice (LV [Ryzen 3900X], NY (New York) [Xeon E3-1241 v3], LU (Luxembourg) [Xeon E3-1241 v3] with Anycast.
  • The above 3x 1GB/$3.00 Slices plus 1x 8GB/$30 Slice in LV (Ryzen 3900X)
  • 1x 8GB/$30 in NY (Ryzen 3900X).

I currently am with the 8GB/$30 one in NY, to which I came back once I let the 8GB/$30 one in LV go.
My reason for letting it go was that LV is too far away from me, geographically, and both NY and LU did not have Ryzen or Slabs (what they call their mountable block storage) at the time.
When I cancelled, I promised that I would come back the moment either LU or NY would get Ryzen and Slabs. So I did when NY got them. I plan to do the same when LU receives Ryzen and Slabs, with one caveat: better peering is needed (see below).

Overall, I am very satisfied with the service. The VPSs perform greatly. They are snappy for my usage, even the $3.00 ones and even those with the older Xeon CPUs. I host websites, "cloud" sync services, automation services, git, wikis, adblocker, seatable, and so on. I have never had any significant downtime, although some users do complain about downtimes on Discord, so I might just have been lucky (see blamefran.com). Even when I see this happening, Francisco is extremely receptive and quick to find out solutions.

The offered I/O, both for the NVMe and the Slab, is excellent. This is the NVMe at the root partition:

curl -sL yabs.sh | bash -s -- -ig
fio Disk Speed Tests (Mixed R/W 50/50):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 279.43 MB/s  (69.8k) | 2.32 GB/s    (36.3k)
Write      | 280.17 MB/s  (70.0k) | 2.33 GB/s    (36.5k)
Total      | 559.61 MB/s (139.9k) | 4.66 GB/s    (72.8k)
           |                      |                     
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 2.46 GB/s     (4.8k) | 2.43 GB/s     (2.3k)
Write      | 2.59 GB/s     (5.0k) | 2.60 GB/s     (2.5k)
Total      | 5.05 GB/s     (9.8k) | 5.04 GB/s     (4.9k)

Slabs are also cheap, at $1.25/month for 250GB. I wish I could share I/O tests of the plain disk, but I have my Slab fully encrypted (LUKS) so the numbers are lower than they are supposed to be. IMHO, they are very good anyway!

curl -sL yabs.sh | bash -s -- -ig
fio Disk Speed Tests (Mixed R/W 50/50):
---------------------------------
Block Size | 4k            (IOPS) | 64k           (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 43.79 MB/s   (10.9k) | 74.73 MB/s    (1.1k)
Write      | 43.86 MB/s   (10.9k) | 75.12 MB/s    (1.1k)
Total      | 87.65 MB/s   (21.9k) | 149.86 MB/s   (2.3k)
           |                      |                     
Block Size | 512k          (IOPS) | 1m            (IOPS)
  ------   | ---            ----  | ----           ---- 
Read       | 183.94 MB/s    (359) | 219.69 MB/s    (214)
Write      | 193.71 MB/s    (378) | 234.32 MB/s    (228)
Total      | 377.66 MB/s    (737) | 454.02 MB/s    (442)

If you need sequential read/write "dd" tests of the Slab, please let me know. I won't run further benchmarks on the Slice, though, because I do not want to stop my services unless it is necessary. You can find several of them on LET, though.

Their bandwidth is unmetered, yet you can not use their 1gbit/s uplink all the time. Rule of thumb (as far as I remember. I do not use a lot of bandwidth) is 100mbit/s 24/7 for the 4GB/$15 Slice, and you halve/double it going down/up in their pricing. I find it fair, and it exceeds my needs anyway.

On their support mechanisms. This is where BuyVM shines. First, their official ticket support system is useful and rather quick to fix issues. I received twice an IP with bad reputation, and they swapped them within minutes.

Then there is Discord. Despite their fair warning that Discord can cause cancer, as long as you stay away from the "misc" channels (or don't. Who am I to judge), people in there are supportive and non-judgmental. This includes both users and staff. Francisco hangs in there quite often and pretty much answers any inquiry, from issue reports, to feature suggestions and chitchats. You can pretty much see that, beyond the memes, he takes a lot of pride in his products and wants them to just perform the best. He also often asks users for what might be the next best thing to add to BuyVM.

Speaking of which, I find their pace of innovation to be schizophrenic and, at the same time, fair for the price. Latest additions in the last months include Cloudflare magic transit for their DDOS protected IPs, which costs a lot of money but is offered at $3.00 per IP, 5 free snapshots of any Slice and scheduled backups for ridiculously cheap pricing, custom ISOs for installing, copy/paste in the Web console (which is WAY more useful than it sounds), and BGP sessions to announce own networks.

No review should omit points for improvement. And there are, plenty of them. Some of them are my humble opinion and might not apply to most users. In no particular order:

Pricing does not scale up a lot, meaning that the least you pay, the higher is the value, IMHO. Everyone has their breaking point, but to me, the price/performance ratio of their first 3 Slices ($2.00, $3.50, $7.00) is way higher than the one of the three successive Slices ($15.00, $30.00, $45.00). Reason is that, especially after the $30.00/month mark, you start finding dedicated servers with perhaps older CPUs and lower characteristics (e.g., a single SSD). On the other hand, many features are included for zero price increase (e.g., snapshots, anycast) or with really low pricing (e.g., backups, DDOS protected IPs., Slabs). Another issue for some users is that pricing is non-flexible for Slices. For example, you can't have a 8GB ram Slice for less than $30/month if you needed, say, less storage.

Peering and transit could be improved, and it is not balanced between datacenters. LV is nicely connected (but too far away from where I am), followed by NY and then LU. The latter seems to be using Cogent only, which does not work very well in Europe. I was getting better transfer speeds from LV to Europe than from LU to..Europe. And I do not even live that far away from Luxembourg.
Speaking of LU, I feel that the datacenter is a bit of a second class citizen. This even taking into account the corona issues (right now, it is completely understandable that there is little going on there, as BuyVM people are not located in Europe). I am dying to have my data in LU, but this will only happen when LU receives Ryzen (and I heard that this might be coming by the end of the year), NVMe, Slabs, and better connectivity.

Edit: As a side note, NY also seems to use Cogent for Europe, but for some reason, it performs well, whereas LU does not. Maybe it is a matter of congestion at peak times.

The "single core" speed of Ryzen is good, but underwhelming on benchmarks (I see scores around 800-900 for Geekbench 5, whereas it is more typical to see scores around 1200-1300). Of course I brought this to the attention of Francisco, who transparently told me that he disabled most of AMD 4ghz+ turbo burst mode (sorry, I have no idea how AMD calls it) because of thermal throttling issues. On the one hand, this looks underwhelming, especially seeing bench results shared on LET/LES. On the other hand, two things: first, the VPS has never let me down. It just flies. I would have not noticed underwhelming CPU performance if I did not run benchmarks. Second, I do not know to which extent the turbo mode actively works for "normal" usage (normal as opposed to benchmarking stuff), so I do not know if and how other hosts that show higher single thread scores actually provide that performance on a daily basis. It might be better to have predictable good performance rather than mediocre performance with spikes of excellent performance. Nevertheless, Francisco promised me he would look at it and consider tweaking some settings.

Final one, and I will just say it: I find their overall ordering process lacking in many ways, potentially irritating for current customers, too. Having to wait for the 1st or the 7th of the month to snatch a VPS sucks, especially when there is a need to scale or offload. To be fair, this issue is almost non-existent currently for NY, because of the new nodes. Related to this issue is when their anti-fraud system kicks in. I completely understand their needs to protect themselves, but it is infuriating to wait weeks before finding the Slice you really needed, purchasing it, and find out that the order got rejected, even if you are a current customer who pays their bills, just because you forgot to disable your VPN, or whatever. Even if you open a ticket immediately after the issue shows up, their response will be that you need to order from scratch. And this might mean that you need to wait weeks. Again.
I think that existing customers should somehow get whitelisted.

Overall, I highly recommend the service and to at least try out their $3.00/month Slice and a Slab. They might do wonders. I also recommend trying out BuyVM for their exceptional support and their community. I am fairly sure that when Francisco reads this, the pride he takes in his service will result in some improvement one day or another.

Feel free to ask further questions about my experience!

Comments

  • That is kind one hell of positive review. I also have a good experience with them. BuyVM is very premm..

    A simple uptime dashboard using UptimeRobot API https://upy.duo.ovh
    Currently using VPS from BuyVM, GreenCloudVPS, Gullo's, Hetzner, HostHatch, InceptionHosting, LetBox, MaxKVM, MrVM, VirMach.

  • How are you using the anycast? I'd hoped to hear more about that in the review.

    The Ryzen cpu isn't spectacularly fast per core, but having so many cores helps smooth out cpu availability fluctuations from noisy neighbors, compared to the old E3's that it replaced.

  • FranciscoFrancisco Hosting ProviderOG

    @willie said: The Ryzen cpu isn't spectacularly fast per core, but having so many cores helps smooth out cpu availability fluctuations from noisy neighbors, compared to the old E3's that it replaced.

    Thanks for the review :)

    We've been doing some tweaks to our Ryzen's to help with heat. If you allow it to go all-core 4Ghz+ it'll hit 95C quickly and thermal throttle itself into the ground. I've been meaning to test out ondemand with the newer kernels we're rolling out but we only recently have we been able to run 5.4+.

    Luxembourg is a tough one. Our plan was to roll Ryzen to all locations this year (and even did the purchases to do so) but Rona's ruined that. NY just recently got upgraded though so there's still hope for LUX to get upgraded this year :) We're working on getting at least 1 more bandwidth provider over there but things have moved even slower over there.

    Always working to better the product :)

    Francisco

  • Being European I've been waiting for slabs to be available in LU to move to BuyVM, so the part about bad peering and transit in LU worries me. Cogent sucks in Europe so if they only use that it does not surprise me that it's bad, but according to the website they have TeliaSonera too which is prem here.
    Any users with vps in LU that can give some opinions about the transit there?

    Thanked by (1)kkrajk
  • @rcy026 said:
    Any users with vps in LU that can give some opinions about the transit there?

    It’s single homed behind Cogent. So it’s very meh.

  • edited September 2020

    @willie said:
    How are you using the anycast? I'd hoped to hear more about that in the review.

    The Ryzen cpu isn't spectacularly fast per core, but having so many cores helps smooth out cpu availability fluctuations from noisy neighbors, compared to the old E3's that it replaced.

    I used anycast to learn about anycast, mostly. I experimented with the following:

    • nameservers.
    • small CDN.
    • a geo-distributed reverse proxy for my homelab to access various services including streaming content.
    • the same as above but with my LV bigger VPS.

    Anycast is where the Discord community helped me the most, now that I remember. I remember realizing I knew nothing good about it and nobody ever judged me for this.

    @rcy026 said:
    Being European I've been waiting for slabs to be available in LU to move to BuyVM, so the part about bad peering and transit in LU worries me. Cogent sucks in Europe so if they only use that it does not surprise me that it's bad, but according to the website they have TeliaSonera too which is prem here.
    Any users with vps in LU that can give some opinions about the transit there?

    In case you didn’t know, they have a looking glass you can try: https://speedtest.lu.buyvm.net/

    I have tried various mtr tests when I had the VPS, to various European IPs, datacenter as well as residential. I have never seen anything but Cogent, so I tend to believe @debaser here.

    @Francisco said:

    @willie said: The Ryzen cpu isn't spectacularly fast per core, but having so many cores helps smooth out cpu availability fluctuations from noisy neighbors, compared to the old E3's that it replaced.

    Thanks for the review :)

    We've been doing some tweaks to our Ryzen's to help with heat. If you allow it to go all-core 4Ghz+ it'll hit 95C quickly and thermal throttle itself into the ground. I've been meaning to test out ondemand with the newer kernels we're rolling out but we only recently have we been able to run 5.4+.

    Luxembourg is a tough one. Our plan was to roll Ryzen to all locations this year (and even did the purchases to do so) but Rona's ruined that. NY just recently got upgraded though so there's still hope for LUX to get upgraded this year :) We're working on getting at least 1 more bandwidth provider over there but things have moved even slower over there.

    Always working to better the product :)

    Francisco

    Thanks for being awesome, especially when there are issues!

  • debaserdebaser OG
    edited September 2020

    @sgheghele said: I have tried various mtr tests when I had the VPS, to various European IPs, datacenter as well as residential. I have never seen anything but Cogent, so I tend to believe @debaser here.

    Their network page is just outdated. Luxembourg is Cogent-only with an HE.net IPv6 tunnel for full v6 routes.

    It's meh, but depending on what you want to do with the box, it can be sufficient. For webhosting it's more than enough, for a Plex box you might want to look to providers with better connectivity.

    Thanked by (1)vimalware
  • @debaser said: For webhosting it's more than enough

    Unless you care about the belligerent DTAG ? I was often seeing speeds measurable in kb/s with LU. Not really BuyVM fault, but of course they can attempt to improve.

  • Fran if you don't mind my asking, how is cpu utilization looking on the slice nodes now, compared to back with the e3's? I'm on a 1gb slice and am pretty sure the host node got cpu starved a few times before the changeover. It's been nice since then as far as I've noticed, but I don't compute much on it, partly by reflex.

Sign In or Register to comment.