Cache memory

[Peter Zeiszler]

I recently rebooted a machine after applying the latest patches for 7.3-1.
Patches:
VMS731_ACRTL V4.0
VMS731_VMSMUP V1.0
VMS731_RMS V6.0
VMS731_SYSINI V1.0
VMS731_AUDSRV V2.0
VMS731_GRAPHICS V4.0
VMS731_FIBRE_SCSI V7.0
VMS731_LAN V13.0
VMS731_DRIVER V3.0

I am not having cache memory showing allocated of 1.6gig with maximum of 5gig.
It used to allocate the complete 5gig on boot and then decrement as applications (processes) requested more memory. This time though it started with the allocated around that value.

Anyone know what controls that allocated value and why it didn't start with the allocated of 5gig after the reboot?

[Ian Miller.]

what does SHOW MEM/CACHE show you now?

Are you using XFC or VIOC?

[Peter Zeiszler]

XFC. VCC_Flags=2.
Extended File Cache (Time of last reset: 14-DEC-2006 13:05:23.85)
Allocated (GBytes) 1.66 Maximum size (GBytes) 5.00
Free (GBytes) 0.00 Minimum size (GBytes) 0.00
In use (GBytes) 1.66 Percentage Read I/Os 78%
Read hit rate 94% Write hit rate 0%
Read I/O count 9390272 Write I/O count 2609222
Read hit count 8920570 Write hit count 0
Reads bypassing cache 416 Writes bypassing cache 1101952
Files cached open 845 Files cached closed 8617
Vols in Full XFC mode 0 Vols in VIOC Compatible mode 24
Vols in No Caching mode 0 Vols in Perm. No Caching mode 0

[Peter Zeiszler]

Sorry - that formatting really is terrible.

Pretty much showing allocated 1.66g and max 5gig.

I am thinking of creating script to just output that to a file every 30 min or so just to see if it changes over time.

[Steven Schweda]

I don't know anything, but I'm at V7.3-2
(with most if not all current ECOs), and mine
(with 2GB of memory, after being up for
months) says things like:

Allocated (MBytes) 145.50
Maximum size (MBytes) 1024.00

and then, after I do some file I/O:

Allocated (MBytes) 165.87
Maximum size (MBytes) 1024.00

So, I'd guess that it might not allocate
storage until it actually needs it for
something, and it's probably giving it back
when it's idle.

[Hein van den Heuvel · Accepted solution]

XFC should only allocate pages when/if needed.
Right after startup I would expect low usage and if anything, that 1.6GB seems high to me, and surely better than 5GB.

As long as it grabs the 5GB if and when needed, concurrent application competition permitting, then all should be well.
For now, I see no indication anything it wrong. The cache hits are there to prove it!

Now I happened to talk to Mark H last week and I believe he indicated a fix where page file data is no longer cache, and execuatable image data is cached better. Sorry, details are vague.
Maybe your prior startup read a crashdump making it fill the cache early on?

fwiw,
Hein.

[Andy Bustamante]

You can use T4 to collect XFC stats, along with other system metrics. See http://h71000.www7.hp.com/openvms/products/t4/ for the download.

Andy

[Peter Zeiszler]

Actually the last reboot was after a crash. However before that it allocated the 5gig right after boot up and that wasnt a crash.

I definitely have had I/O since boot from normal operations and even backups (netbackup and even a local backup). Wonder if something in those patches cause memory to be handled differently.

[Doug_81]

Peter:

Your cache appears to be working correctly. The important figure is the "Maximum Size" - this is what the cache will grow to if the memory is available and yours is sitting at the full 5GB

Although I don't have the luxury of GB of memory, I try to give the cache as much as possible, so with main memory sitting at 1024.00MB I have VCC_MAX_CACHE set to 768.

User are starting ot logoff the system and the evening batch jobs are starting so show mem/cache reports are now showing:

Allocated (MBytes) 547.06
Maximum size (MBytes) 768.00

After the batch jobs get into full swing, the report changes to :

Allocated (MBytes) 752.30
Maximum size (MBytes) 768.00

Both situations report a 98% Read hit rate which tells me that I could probably benefit from some additional memory, but I know that's not going to happen.

You, on the other hand are showing a 94% Read hit rate with 1.66GB allocated. So, there wouldn't be any benefit in allocating additional memory (at that time).

Doug

[Peter Zeiszler]

Oddly enough on our secondary node (configured just like the primary) after the reboot it shows 4.84 allocated GB with 92% hit rate.

I have monitored the primary and its now at 2.41 GB allocated. So it is slowly allocating space.

I honestly can't tell why this one system is allocating less at startup and ramping up as needed, when it has always been using close to the 5g every other time it boots.

For those not using XFC cache I can tell you that was one of the biggest boosts in performance we have had. The other was migrating from HSJ disk controllers to Fiber disk array (XP1024). That move was a huge improvement on our I/O. So if you get a chance of increasing memory for cache, try it, you might be surprised at the improvements you can get.