[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [leafnode-list] 2.0b8_ma9 resource usage on P-150

Stroller schrieb am Freitag, den 08. Februar 2002:

> I assumed that was the case - your version appears to spawn extra threads, or 
> fork 5 child processes, or something. Presumably this makes more efficient 
> downloading, but also for a little extra work for an old system like this - 
> it has to write articles to disk faster, "5 at a time". Internet connection 
> has is 512kbps ADSL, which is new to me.

No, the version does not spawn new threads or child processes. In fact,
there are two situations when forking occurs at the moment:

1. when a new article is posted via nntpd, a background process tries to
   store the article in the spool. When some other process like
   fetchnews or texpire is running, the article is instead placed in
   $spooldir/in.coming/ and stored into the spool later.

2. fetchnews forks a single child when it has finished downloading all
   articles, which then puts itself into a new session, so the fetchnews
   originally started detaches from your tty and exits, and the child
   runs independently. This allows you to use constructs like

   killall pppd

   in /etc/ppp/ip-up.local or something like that.

> It is local to the P150 on which leafnode runs, on a 2gig EIDE hard-drive. I 
> can't recall the exact specification of the h/d, but it was brand new (from a 
> warranty return) about 2.5 years ago. It is probably of average specification 
> for the time the original drive was bought, in 1997.

I have an old Western Digital Caviar AC31200F (1.2 GB EIDE, PIO3 drive)

> I was considering updating the harddrive in the forseeable future, to 
> something like an 80gig or 100gig drive, for spool/news & network filespace 
> (DVD -> DivX rendering can use 20gig per disk using "FairUse"). Since I'm 
> fairly new to Linux & not up to date on hardware, I'd be grateful for any 
> comments - is a 7200rpm drive worth the extra money..? Would I be advised to 
> upgrade to a 2nd-hand Celeron or PII & m/b at the same time..?

Some things to consider: Some IDE chips in Pentium-150 machines don't
speak UDMA/2 (aka UDMA/33). You'd most certainly want to make sure your
chip talks UDMA/33 or consider buying another motherboard or UDMA to PCI

Be prepared to install a "Drive Overlay" into your master boot record
when installing the new drive, old BIOSes (Award 4.51PG in particular)
tend to lock up when they drives with capacities beyond 32 (binary)
GBytes. That may require that you re-install your OS.

Whether your drive spins at 5400/min or 7200/min, does not tell anything
about its actual speed. There are 5400/min drives which are actually
faster than 7200/min drives, some Maxtor 5400/min drives are claimed to
be very silent yet blazingly fast, but I don't have those myself. Also,
7200/min drives can be slow or fast compared against each other, I bet
my "old" (yr. 2000) Fujitsu MAH (7200/min U160-SCSI drive) still
outperforms most current 7200/min IDE drives when it comes to average
latency. If you're after real fast drives, get a good 10.025/min
U160-SCSI drive, but they are expensive.

The German c't magazine has rather useful tests by their "h2bench"
benchmark tool (however it reports bogus results for drives with caches
beyond 4 MB), so if you have model numbers, I can look how their h2bench
results are.

OTOH, 73 GB U160-SCSI drives such as the Maxtor Atlas 10K III or the
Seagate Cheetah 36ES (NOT X15) are quite expensive and may not be an
option, and an Ultra SCSI160 adaptor costs another 250 EUR in Germany.
When configuring a machine today, I'd probably go for a small (say 18 or
36 GB) U160-SCSI drive for the system and a big ATA drive as data store.
However, also consider what backup technology or drive to buy.

Talking of IBM DTLA: keep your hands off those, the MTBF I got for the
seven DTLA 307045 in my vicinity is as low as 35,000 hours: 2 out of 7
failed 9 months after initial operation. Not sure if IBM IC35LxxxAVERxx
are better, but it seems as though a lot of DTLA-307075 were still on
sale. Leave them where they are, don't buy them.

> I then ran fetchnews, so it d/loaded *all* my regular groups from the server 
> fresh again. I susbscribed knode to 1/2 of them, hit retrieve, & groups 
> were available in seconds. I then went out for a while, and when I came back 
> subscribed the the remainder of the groups with kNode. I hit retrieve, and it 
> started getting articles; when I ran fetchnews it slowed, then timed out 
> about a minute later. After leafnode wrote 
> /var/spool/news/uk/telecom/broadband/.overview, kNode was able to sync again.
> freshnews was formerly running with an update frequency of c 10 minutes - 1.9 
> was fine like this, but I regret to say this made 2.0b8_ma9 unusable.

Is anything in the logs? Under certain circumstances, leafnode rereads
its active file all over again on every access, but that's logged as
warning usually (this affects all leafnode programs). Reading news why
fetchnews is running may also be very slow because the server has to
re-read the active and overview files again and again. That's a known
problem at the moment; however, usually, fetchnews finishes quicker than
when catching up with gaping holes in its news store.

Is a local.groups file present in /etc/leafnode (it can be empty)? Is a
/var/spool/news/leaf.node/groupinfo file in place?

leafnode-list@xxxxxxxxxxxxxxxxxxxxxxxxxxxx -- mailing list for leafnode
To unsubscribe, send mail with "unsubscribe" in the subject to the list