Need an opinion: for a server I want to build where the 4 users would have less then 10GB of Data, is there anything wrong with just using 2x 16GB USB sticks in a RAID1 on Linux?
I remember interning at a place where an engineer had a workstation with an APC battery backup and he said he had like 833 MHz processor and 128 MB of RAM and I was like "...................WOW"
I'm about to ask a question about how to assess the best performing server to ssh into given ssh access to 30+ shared computers. Do you guys think this question should go on Super User, Server Fault, or another Stack Exchange site?
it should be fine to ask on SU though, it isn't a "code" question (clearly SO) or an "enterprise iron" question (clearly SF)
there are a great many questions that fall within the spectrum of topicality for all three of the trio sites, although the degree to which it falls in ranges from "dubious" to "extremely relevant"; I think this is one of those cases
your question would probably be dubious (but not closed, probably) on SO; pretty much accepted on SF; and quite nicely topical for SU
not enough enterprisiness for SF :P
unless you mention, I don't know, that each SSH server is running on a VMware vCenter guest... THEN you have their attention
the concept that perks me up a little bit is that your question is about performance -- we care a great deal about performance on SU, or well, many of the regulars do
because one level of virtualization isn't slow enough
nested hardware-accelerated VT-x/EPT doesn't add too much overhead, but if you're using software virtualization inside a paravirtualized or hypervised guest, prepare for pain
I'm writing a question about shared machine usage and talking about how "strained" the machine is (as in how much of its resources are already being taken up by others)
load factor is generally used in reference to the CPU (not so much RAM, network, I/O, etc, although it can be used there too) to describe what percentage of the resources are used, or what multiplier beyond the total resource usage is being contended for
the load factor from top works like this: if you have a 1-CPU system (no hyperthreading and 1 core), a load factor of 0.50 means you're using half your CPU resources; 1.0 means you're using it all; and something greater than 1 means that you have tasks demanding more CPU than is available, so they are therefore running at less than optimal speed
if you have more than 1 hardware thread (due to cores, processors, or hyperthreading), the 100% utilization number increases to the number of total threads you have
so if you have a box with two CPUs with six hyperthreaded cores each, 24 total threads, only a load factor of 24.0 would be total utilization
greater than 24.0 in that case would indicate contention for more resources than are available
@JourneymanGeek his question is about how to find the fastest node for SSH out of a system of 30+ boxen... I don't know if that's quite the same as capacity planning
@Backus the fastest responder will generally have an empty (or at least relatively shorter) queue/buffer on the networking stack, which would definitely indicate that it is either processing packets faster, or fewer are coming in
if the files that the SSH daemon needs to access to authenticate you are all cached in RAM (page cache), you won't hit the disk layer at all by just logging in, probably, which would mean that a simple time ssh would measure CPU and networking, but not I/O load
and the number of users might not be indicative of load if half of them are just sitting there going "hmmmmm how do I fix this compile error" :P
a box with two or three students running multithreaded programs pegging as many CPUs as they can get their hands on, is going to be busier than a box with 15 students using emacs
@Boris_yo @CanadianLuke It's a 60 GB SSD I plan to use for swap, actually. 32 GB RAM installed which means that with proper swappiness settings in the kernel, I probably won't hit swap at all except under truly exceptional circumstances, but I also use ZFS which can be quite memory-hungry and does not play well at all with the out-of-memory killer. Also I'd like to implement hibernation later, which in Linux uses swap space rather than as in Windows a separate file on disk.
@Boris_yo No, I looked back through the transcript but don't think I saw such a link. Would you mind posting it again?
I know putting heavily used swap space on a SSD isn't the best of ideas in terms of device longevity, but like I said, in my case I likely won't hit it much at all, if indeed at all to begin with (before enabling hibernation). I view swap space as a safety buffer, not a replacement for RAM.
Yeah, like this one: "Pagefiles are a relic of the past, dating back to the times when memory was scarce, only 2, 4 or maybe 8 GB memory" :D The first PC I used to do much serious stuff on had 4 MB RAM, later upgraded to a whopping 8 MB. Stuff ran at an amazing speed with that upgrade.
I honestly don't recall how much RAM the first PC I used at all had, but it was a 286, which sort of dates it.
@JourneymanGeek That's roughly the way I figure it too. While the drive itself might wear out, if a part of "RAM" becomes read-only the likely result would seem to be a system crash because of invalid state of some data structures. And with 32 GB RAM and a swappiness value of 10, well... let's just say I don't really hit swap a lot even now.
Plus, I'd much rather stress a swap device a little more and even if only by a slim margin reduce the likelihood of failure of the system drive. Backups run automatically every 24 hours so even with a total failure I don't lose much (especially since system files don't change much), but I don't fancy rushing out to buy new hardware and restoring from backup either.
For such use, a cheaper (in absolute terms), smaller SSD would seem almost ideal. And yes, the idea of putting ZIL and L2ARC on one has crossed my mind as well.
If anything, ZFS would be the write hog on such a combined drive.
(just to put it out there anything I know of ZFS is via osmosis from other people talking about it or wikipedia. It might be, for all I know, a horrible idea ;p)
250gb samsung 840 (cheap and decent) 3tb HGTS (which backblaze seems to love) 40gb fujitsu (for linux, cost me a tenner) and a 250gb hard drive with bad sectors for ephermal storage (trying to kill the thing)
at some point I'm pondering getting a consumer nas, or building one off an old atom box I have lying around
@JourneymanGeek I've thought about that as well, but it wouldn't really save me much because I'm almost always on my main PC anyway. Which has the heavy storage.
The main problem for me right now is that the system drive appears to be dying. But really, after a service life of over 42000 hours (four years nine months), I'm not complaining. I'm pretty sure it came with a five years warranty but I don't think I'm going to bother.
I woke up this morning to a notification email with some rather disturbing system log entries.
Dec 2 04:27:01 yeono kernel: [459438.816058] ata2.00: exception Emask 0x0 SAct 0xf SErr 0x0 action 0x6 frozen
Dec 2 04:27:01 yeono kernel: [459438.816071] ata2.00: failed command: WRITE FPDMA QUEUED
...
The plan is to add redundancy to the main storage ZFS pool too. Right now it's only a single drive; I plan to add one more and run a mirror pair before doing much of anything else to the storage solution.
I wasn't planning on replacing the system drive now but...
As long as the screwup doesn't involve the "destroy" operation. :P
All things considered, I'd probably rather have redundancy than no redundancy, now that the system offers the capability of both redundancy as well as using it to silently recover from (recoverable) errors.
@MichaelKjörling @JourneymanGeek With SSD becoming read-only and 32GB of RAM you won't encounter system crash but will lose unsaved data (which you could not save anyway on read-only SSD) and may not boot properly or at all if critical data could not be saved on last shutdown.
@Boris_yo Um... wasn't the issue the swap earlier? Swap is supposed to be ephermal. The data files are backed up so worst case I'd lose the last day's worth of changes to the OS and configuration. I think I could live with that.
@MichaelKjörling When SSD becomes read-only, not only is it problem with swap if at all, but any data that was crucial to be written before shut down was made.
@Boris_yo So it will behave roughly (not exactly) as if I'd done a forced power off. Not nice, but not a huge problem. If it's really a problem, I'll just restore from the most recent backup. I'd need a new drive at that point anyway.
Everything critical is on the ZFS pool anyway, where it's protected by ZFS' copy-on-write behavior, multiple layers of checksumming and for the really critical data multiple redundant copies.
> With regards to the server specs, we will be looking at a server with just over 1Tb of storage, 32Gb memory, dual quad core processors, redundant power supplies, raid controller, with options on backup drives.
@_@
this is to replace a 9-year-old server...
"overkill" mean anything to these guys?
@MichaelKjörling Swap is used to improve performance even with an abundance of memory.
Swapping out something that is rarely accessed frees up RAM for caching other files.
@Bob I rebooted the PC last night actually so the numbers aren't completely representative, but right now it's sitting at 17 GB RAM free and of the used, buffers+cache is a little over 1 GB.
I woke up this morning to a notification email with some rather disturbing system log entries.
Dec 2 04:27:01 yeono kernel: [459438.816058] ata2.00: exception Emask 0x0 SAct 0xf SErr 0x0 action 0x6 frozen
Dec 2 04:27:01 yeono kernel: [459438.816071] ata2.00: failed command: WRITE FPDMA QUEUED
...
