HSG80 specs

[Wim Van den Wyngaert]

Does anyone knows where I can find the specs of HSG80 ? I'm thinking of how many IO/sec, thruput etc and this also for dual controller config.

Wim

[Jan van den Ende]

Wim,

maybe you better ask that in the SAN forum...


Proost.

Have one on me.

Jan

[Uwe Zessin · Accepted solution]

Some data is in the specifications:
http://h18000.www1.hp.com/products/quickspecs/10545_div/10545_div.html

It says: 'over 24K IOPs per controller pair'. Max through a pair of controllers is about 50 to 60 MegaBytes, if I recall correctly. Max writes to a disk mirror I have seen is about 10 to 15 MegaBytes.

[Wim Van den Wyngaert]

Anyone an idea what the number of write IO/sec could be when the cache is full (sustained heavy load) ?

I see 5000 writes with blocksize 4 pages, thus 10 MB thruput.

Wim

[John Koska]

I believe http://h18006.www1.hp.com/support/techpubs/installation_guides/ek-sma34-an-a01.html will help you. View or download the PDF.

The application note speaks to the performance limits of various FC-Fabric components (Dual HSG-80 controllers performance of cached data and direct media in both IO/sec and MB/sec) for OpenVMS in Table 1.

:) jck

[Robert Atkinson]

Wim, we run dual HSG's with full mirrored cache, pretty much the setup you're talking about.

I've seen IO rates in 'MONI DISK' up to about 8000.

We currently have 4 shelves per pair, two in single bus mode, and two in split bus mode. Almost all of the bays are full (50+ disks), and we haven't seen the controllers flooding yet, even running 4 simultaneous data backups at the same time as doing a Dayend!

Rob.

[Wim Van den Wyngaert]

John, sorry but can't openb the PDF. Will try at home.

Robert,

We have peaks of 6000 WRITES and 1000 READS.
But all IO is done on 2 sites via FDDI shadowing, so there is an additional delay.

Did you look at the queue length ? And I don't mean asking 24 hours in TNG PA but asking an interval of 20 minutes or less. Then you see the real peaks (otherwise they are averaged in a bigger time interval, which is already 2 minutes in my test case).

Wim

[Robert Atkinson]

We have odd peaks at a queue length of 20, but most of the time it's between 0 and 2.

Interval is 5 minute samples.

Rob.

[Wim Van den Wyngaert]

With 5 minutes it is normal. We see only 1 peak in an interval of 2 minutes.

Wim

[Wim Van den Wyngaert]

As I read the data :

1 disk can do maximum 12K IO because it passes thru 1 controller (thus 50% of 24K). But since most IOs contain more than 1 block, the controller thruput of 20 MB/sec is limitting the number of IO's (e.g. to about 2500 when IO size is 8K).

Has anyone a dual HSG in which you see thruput of near 40MB/sec or 1 disk thruput near 20 MB ?

Is a fiberchannel switch a limiting thing ?

Wim

[Robert Atkinson]

Wim, as there are so many other factors involved, you can't really use the factory specs.

If you're looking to move from one controller type to another, the best way to do this is compare the specs for both, i.e. HSG80 will be 4 times faster than HSZ50.

You can then directly relate it to how your system runs now and how it 'might' run after an upgrade.

Also, don't expect twice the performance from 2 controllers instead of one. There's a lot of communication between the controllers so you probably only get 70% out of them.

Rob.

[Wim Van den Wyngaert]

Robert,

I think both controllers exchange all write IO info. So I guess that 1 controller can do 50%. Of course when you run on 1 controller only (other down), this overhead will disappear. In case of read IO it could be faster because the write/read data must not be exchanged.

Wim

[Uwe Zessin]

A single physical disk drive cannot do 12K IO/sec, that is the max IO rate of a single HSG80 using special 'benchmarks'. Ken Bates, storage expert from DEC/Compaq/HP mentioned on the last ENSA@Work conference a value of 200 to 250 IO/sec for a modern 15 kRPM high performance disk drive. A single HSG80 can do a bit more than 20 MegaBytes/sec. The problem is that these numbers depend on the number of disk drives and the I/O pattern.

Usually, fibre channel switches are not the 'limiting thing'. Many models are built using a full, non-blocking 'cross-bar' architecture. Others have a built-in 'overcommitment' - that means, for example, 4 ports with 2 GigaBit/sec go to a single ASIC and that one connects via a link < 8 Gigaite/sec to the switch's backplane. Again, the exact configuration and the IO patterns define if this is a bottleneck.

Indeed, in mirrored cache mode, both controllers update their own and their partner's cache memory. As the cache module is connected via a small, old PCI bus (1997/1998 technology) it's clear that you cannot get full fibre channel speed via all ports.

[Jan van den Ende]

Wim,

Robert,

From your Forum Profile:


I have assigned points to 482 of 681 responses to my questions.


Maybe you can find some time to do some assigning?

Mind, I do NOT say you necessarily need to give lots of points. It is fully up to _YOU_ to decide how many. If you consider an answer is not deserving any points, you can also assign 0 ( = zero ) points, and then that answer will no longer be counted as unassigned.
Consider, that every poster took at least the trouble of posting for you!

To easily find your streams with unassigned points, click your own name somewhere.
This will bring up your profile.
Near the bottom of that page, under the caption â My Question(s)â you will find â questions or topics with unassigned points â Clicking that will give all, and only, your questions that still have unassigned postings.

Thanks on behalf of your Forum colleagues.

PS. â nothing personal in this. I try to post it to everyone with this kind of assignment ratio in this forum.


Proost.

Have o

[Wim Van den Wyngaert]

Uwe,

On a mirrored disk without interbuilding shadowing I get 300 reads and 200 writes during 8 minutes simply with a copy. So I guess that 1000 IOs must be possible. And the disk I was taking about is a RAID-5, so IO is spread.

Wim

[Uwe Zessin]

Wim,
I don't want to get into an 'I said' / 'you said' series again, but I thought you were talking about a physical disk drive, not a HSG unit with multiple physical disk drives.

It is indeed possible to get more than 250 I/Os out of a single physical disk drive if the requests have a high enough locality so that the head assembly doesn't do half or full strokes (not even speaking about the storage array's cache), but that is not how one sizes a storage system.

Performance on a HSG80 RAID-5 storage set depends on your read/write ratio and whether you are using small random I/Os or large sequential I/Os on writes, because the ACS has optimizations for the second case.

[Cass Witkowski]

Wim,

The throughput depends on whether you want to have lot's of I/O per second or lots of MB per second.

Our application needed lots of 1KB I/O in random areas on the disks. I set up a HSG80 controller pair with 36 9GB 15K disk all JBOD. With read cache turned off and doing random reads of 1KB in size I would max out at about 4,700 I/Os per second. I know that the spec ratings for the controller for accessing cache is a lot higher like 24,000 I/Os per second.

Hope this helps

Cass