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
I did not give you all the insight info so you or anyone to feel "bad", that is not the point.
I shared that info so you all can understand our decisions and the decisions of other Providers that either raised prices on recurring and on new products ore EOL'ed some of the 2023-24 era Promo Offers.
Ultimately it will be the decision of each individual to cope / understand or ignore current reality and continue the mocking.
Cheers!
True, Switching wise, you can get some Arista DCS 32x100GBPS dirt cheap these days, same for 32-64GBPS FC Switches from brocade.
Routing usually hovers at a ±2.5USD for an ASR1001-HX and the usual 9 to 12K USD for an MX204 if you wish to go above 4x10GBPS and more then 2 Full BGP tables.
What will kill you is RAM, DRIVES and NVME.
Also GPU if you wish to do any type of GPU related services.
As for ENTP services, Licensing is the killer, not the 37K price tag on a new Nexus switch
.
It really depends on the end product and your end type of customer.
Yup the prices on new memory and disks are brutal, even hyperscalers are having a hard time buying enough disks.
One can only hope the AI bubble burst sooner than later (cope)
I find this discussion very interesting because another fairly large provider who advertises here has recently (in the last two years) gone in the opposite direction. They revamped all of their storage nodes to use software RAID-10 instead of hardware RAID-60. Apparently, they had enough disasters with the hardware-backed arrays that they decided smaller, mdraid-managed arrays was the right way to go, despite the considerable extra cost of hard drives.
Always fascinating to see different opinions on this topic about what works best for each provider.
as always is all about greed, thanks to AI we can see who is the the real king and loser.
Hardware RAID also depends heavily on the controller, certainly, not all RAID cards are created equal.
I would definitely avoid parity RAID on any controller with less than 8 GB of cache and a properly functioning BBU or flash-backed cache.
And I also think that the next products will be based on HW RAID10.
As the Drive prices today are outrageous, I might as well do the best performance wise for local storage, RAID 10 would be the best performance, reliability, and rebuild times wise.
Forgot to answer this.
Since your Promos were changed for recurring to List priced products, you inherit that particular list priced product default settings.
White Neptune and IO 200 have 1 backup slot as default.
So enjoy your extra features, till your canceled renewal date
What are some of the best, in your opinion?
Depends on you Brand of server
for HP , whatever above HP SR932i-p 8GB, yet the AR series are also good, unfortunately thoes have 2-4GB cache, HP is conservative on their RAID cards.
for DELL, whatever is above H740P, that already has 8GB cache
As for Supermicro, have no clue, yet they use LSI based cards, so either LSI or LSI branded like DELL, ServeRAID from IBM or Fujitsu-Siemens cards.
For HBA, whatever SAS 24/12/6 GBPS card would do, the more modern the better, but I would not use old LSI 2008 Chipset based cards anymore.
LSI is all I've used, and I hated it. I was only happy after switching to mdraid (although I didn't have nearly enough disks for the software RAID CPU overhead to matter).
All these exceptions make it very confusing, just my 2 cents.
The latency problems I am referring to appear under high IOPS and at a large disk count.
Let me expand on this and our findings a bit:
Context matters here. Some people consider a pool of 8–12 disks large, but that is still relatively small compared with the systems I am talking about.
We encountered serious latency issues with ZFS pools containing as many as 96 disks or above. Even systems with 1 TB of RAM did not resolve them, because additional RAM cannot compensate for an unsuitable pool topology or a bottleneck elsewhere in the storage path, or a god damn lazy disk
( that lazy disk will pop up as faulty in a HW raid config or in a branded storage solution ), while in ZFS if you don't monitor it, you will never know.
A properly selected SLOG can certainly improve synchronous write latency, but in our particular workloads, neither the SLOG configurations we tested nor further ZFS tuning produced the consistency we needed, nor the numbers we aimed for.
In our testing, a NetApp FAS system consistently outperformed that ZFS configuration. Unfortunately, it also won decisively on power consumption.
Licensing is another consideration: buying a used or entry-level system without the necessary licensed features can make the entire exercise pointless, obsolete or a waist of time, getting a high end model on the other hand....... but you will fish for CISCO drives all day long for spare parts, same for EMC by the way.
Regarding disk shelves, we found the NetApp DS4243 to be one of the most stable options we tested. The NetApp DE6600 and various Dell PowerVault MD systems also performed well. ( aaa, I hate HP so we did not touch their line
)
However, compatibility depends heavily on the exact controller, IOM firmware, HBA, cabling and drive firmware. SATA disks can be particularly troublesome pain in the ass in enterprise SAS shelves, while consumer SSD are not much better. SSDs were not the intended use case for most of the older shelves we tested anyway, nor that was the scope of our tests.
So I am not claiming that hardware RAID or a FAS system is always better than mdraid or ZFS. I am saying that, at approximately ~96 disks under sustained, latency sensitive workloads, our results were very different from what someone might observe with an 8–12-disk server.
The main complication complain from my view would still have been Proxmox’s comparatively awkward handling of shared iSCSI multipathing and its lack of a native clustered datastore filesystem comparable to VMware’s VMFS. Those limitations make traditional active-active SAN architectures considerably less straightforward attractive than they are with VMware.
It can be made to work using LVM on shared iSCSI storage, or you can go for ZFS over ISCSI, or any other exotic setup, but the overall complexity is not something I wish to wake up at 3AM.
Storage is sensitive to latency, that is the killer, so, as much complicated stuff you add between the VM's disk and the underlying storage, the more latency you get and the more problems you are facing on updates.
Due to all above, in my view, Local Storage is the best cost effective solution for the type of services we sell today, or for what the market expects from a chap/decent storage solution.
If you require advanced features such as high availability and uptime guarantees, then we are discussing an entirely different class of product, with a correspondingly different architecture, operational model and specially different price point.
You cannot reasonably expect enterprise grade redundancy and availability from a budget storage service. ??, yet, it seems some do
- take that with a grain of salt please.
Cheers!
That's why we miss @Calin
Well rumor says that he is turning to the Good Side......
So who knows, your wish might come true...........
And for the fellas here who don’t understand why I keep going on about Fibre Channel, here is a simple illustration.
FC was designed from the ground up to carry storage traffic end of story, while iSCSI runs over the full TCP/IP and Ethernet stack.
With FC, the result is usually better latency, but more importantly, it provides consistency under load and predictable performance.
With TCP/IP over Ethernet, many more factors along the transport path can interfere with storage traffic, create queues, and introduce latency or jitter. The switches themselves matter enormously—their architecture, buffering, MTU configuration, QoS, congestion management, and so on. Ethernet switches capable of handling storage traffic properly can also consume considerably more power than comparable FC switches, depending on their speed and configuration.
Secondly, FC gives you a completely separate, dedicated storage fabric that is not shared with your normal Ethernet infrastructure. That separation is a major advantage when an Ethernet switch decides to die leave the chat at 3:00 AM.
Precisely, yet in SOME, and the cost of that is way above what you get from a simple FC setup, yet again, depends on your end desired setup and need.
Right tool for the Right job.
@host_c check my PM when you get a minute
Just check @host_c
No need to reply