SFS (68k) and recycled dir

Hi J?rg! I hope you can tell something about.

I mentioned something strange. I add a partition (DH11) with HDInstTools 10 days ago. The whole last range of the HD. Until now I used it only for recordings with AHIRecord what writes 256kb at once (editable) to the file. That worked fine until now. Suddenly AHIRecord start writing 256kb and then hangs. I can reboot an start AHIRecord what wirts another 256kb file and hangs. On other partition (eg. DH4) it works well. I thought maybe someting is full or think so. Then I delete .recycled dir (about 8GB) on DH11 and tried AHIRecord again. Now it works. It would be a pitty if this happens again.

On my other (nearly full) partition (DH4) with SmartFilesystem 2.3 (28.09.00) I never had such a problem.

Einheit Gr??e Genutzt Frei Voll Fhlr Status Typ Name
DH4: 12,9G 12,3G 624,4M 95% 0 schr.-/lesbar SFS2 Rec


Here the problem partition:

Einheit Gr??e Genutzt Frei Voll Fhlr Status Typ Name
DH11: 27,2G 8,0G 19,3G 29% 0 schr.-/lesbar SFS0 Rec2

SFSquery information for dh11: (SFS Version 1.279)
Start/end-offset : 0x0000000b:d3234000 - 0x00000012:a1f16000 bytes
Device API : NSD (64-bit)
Bytes/block : 512 Total blocks : 57108240
Cache accesses : 30688 Cache misses : 24855 (80%)
Read-ahead cache : 8x 8192 bytes (Copyback)
Flush timeout : act. 20s - inact. 0.5s
Max Name Length : 107
DOS buffers : 150
SFS settings : [RECYCLED]

SFSCheck:
Partition start offset : 0x0000000b:d3234000 End offset : 0x00000012:a1f16000
Surfaces : 1 Blocks/Track : 324
Bytes/Block : 512 Sectors/Block : 1
Total blocks : 57108240
Device interface : NSD (64-bit)

Checking RootBlocks
...okay
Checking AdminSpaceContainers at block 2
...okay
Checking NodeContainers at block 7
...okay
Checking ObjectContainers at block 3
...okay
Checking Bitmap at block 34 (14278 blocks, 4000 bits/bitmap)
...okay

Alas I have to report that I got the same problem again. After deleting some files from .recycled it worked again.

@daxb

"Cache accesses : 30688 Cache misses : 24855 (80%)
Read-ahead cache : 8x 8192 bytes (Copyback)
Flush timeout : act. 20s - inact. 0.5s
Max Name Length : 107
DOS buffers : 150
SFS settings : [RECYCLED]"

150 dosbuffers on a ~30 GB partition? no way!

@Framiga

>150 dosbuffers on a ~30 GB partition? no way!

Do you mean that the amount of 150 dosbuffers are responsible for the problem? I would use a higher amount but I`ve only 32MB available. Thats the reason for the 150 buffer. :(

@daxb

up it at least to 300
i may be wrong but remember reading somewhere it's the
least amount of buffers when you use SFS partitions.

I have a 25GB partition here using it with 400 buffers and
never ever had problems with it

...although, i may probably be wrong ... waitign for joerg to pop in

@Raziel

400 dosbuffers with a 25 GB partition? your using SFS at 40% of its potential!!

try a SFSQuery and look at how many caches missed you get.

@daxb

if you use that partition for HD recording, set the partition blocksize to 1024 with 1000 dosbuffers and without .recycled

Use SFSFormat with the NORECYCLED option

i.e:

SFSFormat DEVICE xxx: NAME foo NORECYCLED

Edit- btw just noticed this bit:

"I mentioned something strange. I add a partition (DH11) with HDInstTools 10 days ago"

have you always used HDInstTools for ALL partitions on that HD?

@Framiga

Quote:

0% - 10% on all of my partitions...so?

@Raziel

"so" you have no enough dosbuffers!

0% is what you should achieve. (look at the 80% of caches missed from the above reported sfsquery)

Of course you have to test with sfsquery, accessing the partition after some minutes.

@Framiga

I can live with 10% cache misses on the upside of having
more RAM for more important tasks...

@Raziel

beeing a filesystem and not a screensaver, i would care about if i were you!

@Framiga

Uhm, why?

It works, even with 99% cache misses it would work, it would
just have to access the drive more often...

@Raziel

efficency, reliability, performance ... have you checked fragmentation onto your SFS partition for example?

@Framiga

yes ... close to nothing after months of usage
by far better than other fs after defragmentation

@Raziel

Quote:What's more important if using only a few KB more for the SFS buffers makes it 10 or even 100 times faster?
10% cache misses is a lot, if you get more than 1% cache misses (after using the partition as usual for some time, don't check it for example directly after (re)booting) SFS is already much slower than it could be.

On classic Amigas with only a few MB RAM using enough buffers may not be possible, but on the AmigaOne, or a CSPPC with 128 MB or a BPPC with 256 MB RAM, it's not, you usually only need a few KB more RAM for the buffers of each SFS partition to get much faster speeds, not several MB.

@daxb

Quote:It shouldn't, it's only much slower than it could be with that few buffers.

Quote:Using for example 300 buffers should make it much faster and only needs 75 KB more RAM.

@joerg

"@Framiga

>150 dosbuffers on a ~30 GB partition? no way!

Do you mean that the amount of 150 dosbuffers are responsible for the problem?


It shouldn't, it's only much slower than it could be with that few buffers."

nope, not necessarly. I was poiting out that beeing the user using a misconfigured SFS for HD recording (with recycled) the dosbuffers could be part of the problem.

Or more simply he have an HD full of bad blocks.

btw ... do you thik is a good idea using SFS with a slighty larger blocksise (1024-2048) for HD recording? (still without .recycled)

About Raziel i still don't uderstand if he is speaking about AOS4 on AOne or Classic!

@joerg

Quote:

Understood, though you don't say what would be a good
buffers setting for, lets say
256 MB
500 MB
1 GB
5 GB
10 GB
25 GB
70 GB
partitions...(that being my partition sizes)

@Framiga
Quote:

Check my extra info, i try to port stuff, thus needing every
byte of RAM i can get (until VMEM pops up )

@Raziel

lol! you have "problems" with half gig of RAM? i can't beleive!!!

@Raziel

Quote:It doesn't depend on the partition size but on the number and sizes of the files and directories, how they are used, etc.
The only way is to check the cache misses with SFSQuery after using the partitions as usual for some time, if you get more than 1% cache misses you should use more buffers.