disk utilization at 100%

[Lynn Hsu_2]

I have a busy database server (HP-UX 11.11, 2-CPU, 4GB Ram, rp5470) uses EMC Cx500 for storage. There is one LUN shows peak disk usage 100% all the time from gpm(glancePlus). This LUN is on Raid5 stripped across 5 disks on the CX500. THis Lun is for warm spare databases and gets updated every 20 mins. So there is always disk IO (mostly write).There are other LUNs on the same Raid group shows not much disk usage. Also, host side uses static buffer cache 144MB.

The disk subsystem queue is below 1% and memory queue is 0 all the time. I am thinking to change to dynamic buffer cahce at 250 MB to 500 MB. Will this reduce disk % on that LUN? Any suggestion?

THanks in advance,

Lynn

[A. Clay Stephenson · Accepted solution]

250MiB is a small buffer cache so increasing to 500MiB would be wise as long as the box is not under memory pressure. 4GiB of memory is generally considered to be a small amount of memory theses days. In any event, don't be overly concerned when Glance displays 100% disk usage for a disk array LUN. Glance (or any host-based performance tool) simply sees what it thinks is 1 physical disk (eventhough the underlying LUN may be comprised of many disks). All Glance knows is that a bunch of i/o is going to and from this "disk".

[Lynn Hsu_2]

During peak hours, sar rcache% is 74, wcache% is 86. The OS buffer cache is 144MB. We use Progress database SW. I saw
somewhere suggesting bypass the OS buffer cache for the database filesystem. So it looks like we don't need to worry about increasing the OS buffer cache at all?

Lynn

[A. Clay Stephenson]

Even if Progress is using raw i/o for everything there is still ordinary UNIX file activity that does use fully cooked files and benefit from the buffer cache.

[sathish kannan]

Hi Lynnn,

If it is most write as you in your case, why don't you consider using RAI1D 1 ratger tha RAID 5. RAID 1 is faster in write that RAID5.

How big is your disk? How about the disk in the CALRiiON RAID group?Are they busy?


Regards
Sathish

[avik]

I also wondered if there is a higher rate of deletion etc ?

[Lynn Hsu_2]

The busy LUN is mostly write (add/delete records). Other LUNS across over the same set of EMC disks does not show busy I/O. I wonder if we can ignore the disk % from GlancePlus? I was told by HP to ignore disk % since it is disk array. Any comments?

Lynn

[A. Clay Stephenson]

Same comments as before. Glance doesn't have a clue that this is not a single physical disk. Glance's metrics are based on the assumption that it is a single conventional disk. Now if you want to make things APPEAR to be better then the VG into smaller LUN's and stripe across them. Glance will be happier but the overall i/o permformance will likely change very little.

[Lynn Hsu_2]

Thanks!