home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!spool.mu.edu!vms.csd.mu.edu!5916RAHMANK
- From: 5916rahmank@vms.csd.mu.edu
- Newsgroups: comp.sys.ibm.pc.hardware
- Subject: ..SMARTDRV vs Built-in Cache, built-in is faster!....
- Message-ID: <0095F799.3AEAA180@vms.csd.mu.edu>
- Date: 22 Aug 92 18:48:22 GMT
- Article-I.D.: vms.0095F799.3AEAA180
- Sender: news@spool.mu.edu
- Reply-To: 5916rahmank@vms.csd.mu.edu
- Organization: Marquette University - Computer Services
- Lines: 85
- Nntp-Posting-Host: vmsd.csd.mu.edu
-
- In article <0095E1C5.3BDE0260@vms.csd.mu.edu> 5916rahmank@vms.csd.mu.edu writes:
-
- ...
-
- |>
- |> Smartdrv.exe loaded (w/double buffer) Not loaded
- |> ------------------------------------- ----------
- |>1. Time to load win 3.1: 29:00 sec 36:27 sec
- |>
- |>2. Time to load W4W 2.0a: 8:23 sec 12:46 sec
- |>
- |>3. Time to load a 2.11MB file
- |> in W4W. This file has
- |> lots of graphics in it: 3:82 sec 4:10 sec
- |>
- |>4. Time to save the same file: 16:17 sec 19:10 sec
- |>
- |>5. Time to load excel 4.0: 8:29 sec 11:14 sec
- |>
- |> This experiment shows that when smartdrv.exe is loaded it saves time
- |>i.e., the system performs much better than when it is not loaded. This
- |>difference in time (which translates to the performance of the system) becomes
- |>more prominent with systems with lower clock speed. However, the main question
- |>here is smartdrv vs. built in cache. It would be interesting to perform similar
- |>time measurements on other systems and then make a comparison. But still the
- |>question remains: why smartdrv enhances performance even though there is
- |>built-in cache on the drive (as well as on the controller)?
- |
- --------------------------
- |Ah. But how big was the SMARTdrive cache? Something like 2-4MB, I'll bet.
- |The 512K cache on the controller is trivial by comparision.
- |
- |I have a setup similar to the one above. It used to have 512K on the DTC.
- |I tried similar experiments with a 1MB SMARTdrive cache on top of the
- |harware cache. There was a similar significant speed improvement.
- |
- |However, I have finally got 2MB on the controller (what I ordered; they put
- |the wrong SIMMs on the controller :-(... With 2MB hardware cache the system
- |is just as fast as with the 512K hardware cache PLUS the 1-1.5MB SMARTdrive
- |cache.
- |
- |I think the results above speak more about the SIZE of the cache used than
- |the TYPE (hardware or software).
- |
- |The real experiment would be to compare a 2MB SMARTdrive cache against a
- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
- |2MB hardware cache. Unfortunately, I don't think the DTC will run without
- ^^^^^^^^^^^^^^^^^^^
- |RAM installed...
- |
- |The other lesson I learned from all of this fooling around is that defragging
- |your hard disk will provide significant speed increase. At least with the
- |DTC controller, there is software-configurable read-ahead. I have mine set
- |so that if there are two or more requests for data in a linear fashion, the`
- |controller reads the next several tracks into the hardware cache. Result?
- |Windows loads in about 20 seconds. With even a minimally fragmented disk,
- |the read ahead is defeated because there is no way to predict where the
- |next data will need to come from.
- |
- |-Chris newbie@inmet.camb.inmet.com
-
- ==========================================================
-
- Hi Folks:
- Finally I got 4M ram on my DTC SCSI card and I ran the
- smartdrv vs. buitl-in cache experiment again. And yes,
- built-in cache is faster than the smartdrv cache!
-
- (a) Here is one comparison. It took 29:00 sec to load the NDW
- along with lots of icons (win 3.1) when I tested with
- 512K built-in + 1mb smartdrv cache (as I reported before).
-
- (b) With 4mb cache on the card it takes only 12:12 sec! (no smartdrv).
-
- (c) According to previous discussions in this thread, built-in
- + smartdrv should be slower than the built-in cache alone.
- This is precisely what I found, it took 22:00 sec to load
- NDW and everything with the 4mb built-in + 2mb smartdrv!
-
- So now, there is no contradiction between theory and experiment!
-
- These tests were done on a stacked drive, so results may be
- different on an unstacked drive.
-
- - Anis, Dept. of EECE, Marquette University.
-