Instalar Steam
iniciar sesión
|
idioma
简体中文 (Chino simplificado)
繁體中文 (Chino tradicional)
日本語 (Japonés)
한국어 (Coreano)
ไทย (Tailandés)
български (Búlgaro)
Čeština (Checo)
Dansk (Danés)
Deutsch (Alemán)
English (Inglés)
Español - España
Ελληνικά (Griego)
Français (Francés)
Italiano
Bahasa Indonesia (indonesio)
Magyar (Húngaro)
Nederlands (Holandés)
Norsk (Noruego)
Polski (Polaco)
Português (Portugués de Portugal)
Português - Brasil (Portugués - Brasil)
Română (Rumano)
Русский (Ruso)
Suomi (Finés)
Svenska (Sueco)
Türkçe (Turco)
Tiếng Việt (Vietnamita)
Українська (Ucraniano)
Informar de un error de traducción
Raidz2 /raidz3 is a pain when writing to it, though random reads may work faster due to a bit more spindles for reading that have "the same" information (depends on how it's made in software).
I have zvol's, a PC with "main" library, do a snapshot and then serve that snapshot clones to other PC's, main zvol is served to "main" libraru PC.
If things go bad (e.g. the P2 pool runs out of space) - I destroy clones and do a rollback to snapshot
service ctld stop
zfs rollback -rRf P2/MMO256@snap
Well, with modern 60GB titles - it's tens of minutes of waiting (for NAS-based storage) instead of downloading. Not to mention things like SSD wearout for normal desktop use.
Impressive, but I'd check out how cpu's are utilized, and where are all those drives attached to - looks like 4xPCI-E :) ?
raidz2 may have just a bit of extra random reads performance compared to raidz but in writes and linear reads it'll fall behind.
What are the blocksize and recordsize in your zpool/FS (or if you serve zvol's you need to patch libzfs to allow zvol creation with volblocksize>128KB)? Larger recordsize is good for compression and linear read speeds, but absolutely trashes performance in small random reads/writes. In NTFS you end up with 4KB (default) or up to 64KB clusters so the performance will suffer. I had to go (down) to 256KB volblocksize (+I have compression) or my games were taking next to forever to load with 1MB volblocksize.
Having zpool of complete hard drives and not of partitions (=you'll get many pools) will degrade performance a lot - it causes more seeks (which means speed drops down to several MB/s or even under 1MB/s per drive) and hard drive speed is like twice lower at its end. Replacing a drive in a pool will be horrific experience - writing 4TB takes... long. And remember, ZFS is copy-on-write = to change 1 byte in a file on 1MB recordsize pool means to read 1MB (checksum check+ optional decompression), do stuff to get some data to upper layers, change that byte, pass down, optionally compress, calculate checksum, write that 1MB. "read and write amplification", that is.
24 hdd's (or better say 22+2) in raidz2 pool is not the best idea either - with e.g. 4kb blocksize and "default" 128KB recordsize you can't evenly split the load (128kb/22 drives=some float point) and it'll be like 4kb random reads/writes in the end. I'd go with 16+2=18 drives (or raidz3 with 16+3=19 drives). Raidz2 for 24 drives means higher risk of failure - 3 of 24 drives going bad is way more probable than 4 of 24 or 3 of 8 or 16.
Jumping from FreeBSD (FreeNAS) to Linux (Unraid,Arch) kernel is really a significant environment change and I can't predict if it's better or worse at the moment (it's a bit holywarish thing and highly dependent on hardware drivers as well).
In FreeBSD I'd use gstat to monitor drives load to see if the performance is pinned by drives themselves. Check out ARC usage, try turning off ARC compression.
Oh-my :) Wish I could try that. But just a small note - to fill up gigabit read speed (100+MB/s) it was enough 800MHz Ivy Bridge Celeron (2 cores), strangely read speed got lower with higher frequency :). Writes need more CPU power as I use compression.
That is over the net? factor in NIC drivers/performance as well.
Those are 2 different load types - games=random i/o, streams/video editing=mostly linear i/o, where 1MB recordsize or volblocksize will do good.
So, you just should separate some drives for games and some for videos. Not much point in keeping temporary material on raidz2 as well (unless you make $$$ on it and loss will cause financial issues)
Steam supports only like 10 library folders - you should keep that in mind.
And finally - 24 3.5" drives alone will constantly eat over 100W power (peaking at some 300W if they start all at once), it can be more profitable to have local drives than a huge always-on NAS :).
If yopu rpre-allocation is takking hours. Then something is not right with your drive or file system... or perhaps its you AV program mucking things up. Try telling yyou AV to ignore/exclude you steam directories from automatic detections.
Worst idea ever. While Steam does scan things and typical malware must pass that scans to get to you (it's likely it actually will pass your AV as well) - as the time goes by and you have downloaded product on your drive it becomes only your AV responsibility to detect malware...
I can guarantee you. The ANti viurus software steam employs is a league or two what you're running. Plusd if you cactually do incur damages from a virus that came form a gamedistributed through steam, you can sue steam for the damages and then some.
Point is, AV in autoprotect tends to delay all read and write operations. Hence why I recommended creating an exception rule. That will in all liklihood remove those lengthy preallocation cycles.
You somehow think your junky consumer anti virus is going to detect something that sneaks by CloudFire, Akamai, and Steam
Yeah there is a zero % chance of that happening
Thing is that your "my junky av" will detect that virus TOMORROW (or moths later) when it gets the updates, while none of the AV products detect that threat TODAY, when we download and install the app.
Dear Mr. God (cause He is the only one who can guarantee anything),
Noone has a "silver bullet", they can run dozen AV products with all heuristics and such - but as long as they don't get the source code for the apps (and analyze it) - they can only guarantee one thing, their best effort as of "TODAY". Even if they do serious analysis in a sandbox - who says that the trojan or some malware will activate right away as you launch a game? I'm a bit into programming (from BASIC to C and Assembly, and modding FreeBSD/Linux/Solaris kernels) since the end of 1980's - so I clearly see the possibilities that the devs get by installing their software on your HDD. Really endless unless they break something and make the right people really look into the issue. And your gaming PC typically is not "purely for gaming" but contains lots of easily readable information.
Just a couple weeks ago we've seen a "second PUBG" emerging from what was a "Measurement Problem" - a game written in 2016. The devs account was likely hacked and hijacked (no details if the hackers got the game source code or the additional software that gets installed before you run the game for the first time was modified) - resulting in a massive keys giveaway several months ago (we all like giveaways with cards and such - don't we? And we never ever suspect anything wrong in this action) and a strange, meaningless, news article in Russian appearing just that couple of weeks ago with the game changing its title to PUBG (it actually RESEMBLED PUBG due to using similar-scripted letters from the other alphabet). With only the image containing a sentence like "a thousand eyes are watching for you and creating a digital dossier that can be purchased".
The US population is roughly 5% of the world population - that is it'll be from hard&useless to impossible to go to court with such case for most people :). And that's not accounting for the fact you may not be able to even guess and prove the Steam being involved in your real life issues - like theft of documents, identity and alike causing consequences - even if the "consequences" case gets completely investigated it'll end up in dead end like "I purchased this person information somewhere on the net - don't remember where"
A preallocation is technically writing of zero bytes - with RAM speeds at several GB/s AV checks are unlikely to affect that - simply because they stop pretty early as there's not much to compare to zero bytes. Nevertheless, writing to HDD some 60GBytes at a rate of "almost 100MB/s" takes ~10 minutes and doesn't really ensure you'll not run out of space. With trend of using network storage for your games - you definitely will not jump over 125MBytes/s (a gigabit LAN limit, as 10G-100G is not really usable at home so far) and that is rarely needed, in fact you often face random i/o when running games.